refactor: remove the TABLASSERT_FULLMAP_SHARDS build tunable

[SkyeAv]

Removes the TABLASSERT_FULLMAP_SHARDS environment variable so the fullmap RECORDS shard count is fixed at the compile-time cap (16) rather than runtime-configurable. Default builds are byte-for-byte unchanged — 16 was already the default — so this removes only the override, not the sharding.

Build phase

• Removed resolve_shard_count(): the sole reader of TABLASSERT_FULLMAP_SHARDS in rust/src/fullmap.rs; the build call site is now let shard_count = SHARD_COUNT_SHARDS;.
• Kept the shared helpers: round_down_pow2 (still used by the read path) and env_usize (still used by the seven surviving tunables) are untouched — no dead code, clippy clean under -D warnings.

Read-path compatibility

• shard_count_of is preserved: the read path still reads the per-database meta table (shards) and rounds it down to a power of two, so databases built earlier with fewer shards (the former 4-shard layout, or a 2-shard build) still open exactly the shard files that exist and route with the matching mask.
• Tests retained: runtime_shard_count_builds_and_reads_fewer_shards (drives build_fullmap_inner with 2 shards directly) and legacy_four_shard_db_reads_only_four_shards still pass and now document this as the parameterized/legacy path rather than an env-var feature.

Docs

• docs/fullmap.md: dropped the TABLASSERT_FULLMAP_SHARDS row from the Build Tunables table and reworded the Output Artifact shard-file line (count fixed at 16; read path still honors a database's recorded count).
• CONTRIBUTING.md: removed the variable from the example tunables list (the docs/fullmap.md#build-tunables-environment link stays the source of truth).
• CHANGELOG.md: added an ## Unreleased → ### Breaking Changes entry; the two historical 8.0.1 mentions are intentionally retained as the release record.

Design

• Why remove: the shard count is a layout invariant the read path must agree on; exposing it as a build-time knob added surface area (round-down/clamp logic, meta recording, docs) for a value that was effectively always 16. Pinning it simplifies the build without changing any default output.
• Accepted caveat: anyone still setting TABLASSERT_FULLMAP_SHARDS gets it silently ignored (no error); their builds produce the same 16-shard layout as before. Flagged as a Breaking Change in the changelog for visibility.

Testing

• cargo fmt --check --manifest-path rust/Cargo.toml → clean (also enforced by the pre-commit hook).
• cargo clippy --manifest-path rust/Cargo.toml --all-targets -- -D warnings → clean.
• cargo test --manifest-path rust/Cargo.toml → 55 passed (unit) + 10 passed, 1 ignored (golden; ignored = opt-in regenerate_golden), incl. schema_and_shard_count_are_pinned.
• uv run maturin develop --manifest-path rust/Cargo.toml → built + installed the extension.
• uv run ruff check . → all checks passed; uv run ruff format --check . → 63 files already formatted.
• uv run pytest tests/test_fullmap.py tests/test_fullmap_golden.py tests/test_cover_fullmap.py tests/test_rs.py -q --no-cov → 57 passed (incl. the 16-shard layout assertion).
• uv run pytest -q --ignore=tests/test_qc.py → 616 passed, 27 skipped, 1 failed.

Caveats / not introduced here

• The single pytest failure and 8 pyright errors are pre-existing environment gaps, not regressions: both are ModuleNotFoundError for sklearn/numpy in qc.py/test_qc.py, because the [qc] extra (torch/sentence-transformers/scikit-learn) isn't synced in this worktree. No .py files are touched by this PR; CI installs --extra qc (.github/workflows/ci.yml) and runs these green.

Summary by CodeRabbit

• Breaking Changes

  • Fullmap RECORDS files now use a fixed 16-shard layout for newly created databases.
  • The configurable shard-count setting has been removed and is ignored during creation.

• Compatibility

  • Existing databases continue to be read using their recorded shard count, preserving compatibility with older layouts.

• Documentation

  • Updated fullmap configuration and development documentation to reflect the fixed shard layout and removed setting.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 9a30c7b6-fb72-480e-be37-7fa17e1f6161

📥 Commits

Reviewing files that changed from the base of the PR and between ee46534 and c68b8ff.

📒 Files selected for processing (1)

• rust/src/fullmap.rs

🚧 Files skipped from review as they are similar to previous changes (1)

• rust/src/fullmap.rs

---

📝 Walkthrough

Walkthrough

Fullmap builds now use a fixed 16-shard RECORDS layout. Reads continue to use and normalize the shard count recorded in database metadata. Documentation and test descriptions no longer reference the removed environment-controlled setting.

Changes

Fullmap shard layout

Layer / File(s)	Summary
Fixed public shard layout rust/src/fullmap.rs, docs/fullmap.md, CHANGELOG.md, CONTRIBUTING.md	The public build path writes 16 shards and removes TABLASSERT_FULLMAP_SHARDS resolution. Documentation and regression tests describe and verify the fixed layout.
Metadata-based read compatibility rust/src/fullmap.rs	The read path uses META.shards and normalizes non-power-of-two, missing, invalid, and zero values.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• SkyeAv/Tablassert#61: Updates docs/fullmap.md with related Fullmap shard layout documentation.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely identifies removal of the TABLASSERT_FULLMAP_SHARDS build tunable, which is the main change.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch delete-shards-env-var

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (1)

rust/src/fullmap.rs (1)

2240-2242: 🗄️ Data Integrity & Integration | 🔵 Trivial | ⚡ Quick win

Cover the removed environment variable through the public API.

The changed assignment is in build_fullmap_db, but the shown layout tests call build_test, which invokes build_fullmap_inner directly. They cannot catch a regression where TABLASSERT_FULLMAP_SHARDS=2 still affects the public build. Add a test that sets the variable, calls build_fullmap_db, and asserts META.shards == "16" plus all 16 sibling shard files. Restore the process environment in a guard or serialize this test.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@rust/src/fullmap.rs` around lines 2240 - 2242, Extend the fullmap tests to
cover the public build_fullmap_db API: set TABLASSERT_FULLMAP_SHARDS to 2,
invoke build_fullmap_db, and assert META.shards is "16" and all 16 sibling shard
files are created. Restore the environment variable with a guard or serialize
the test to avoid process-wide environment races.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@rust/src/fullmap.rs`:
- Around line 2344-2347: Update shard_count_of and its callers to reject present
META.shards values that are zero, non-power-of-two, or outside the supported
range instead of normalizing them; preserve the missing-metadata default
separately. Remove the test expectation that accepts normalization, and add an
end-to-end lookup test using multiple shard files to verify malformed metadata
cannot silently omit records.

---

Nitpick comments:
In `@rust/src/fullmap.rs`:
- Around line 2240-2242: Extend the fullmap tests to cover the public
build_fullmap_db API: set TABLASSERT_FULLMAP_SHARDS to 2, invoke
build_fullmap_db, and assert META.shards is "16" and all 16 sibling shard files
are created. Restore the environment variable with a guard or serialize the test
to avoid process-wide environment races.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: cab3a110-2b35-4b1d-b18e-1e5d60a3334c

📥 Commits

Reviewing files that changed from the base of the PR and between ee49b62 and ee46534.

📒 Files selected for processing (4)

• CHANGELOG.md
• CONTRIBUTING.md
• docs/fullmap.md
• rust/src/fullmap.rs

[SkyeAv]

Addressed the review:

• Public-API coverage (nitpick) — done. Added public_build_ignores_shards_env_var in c68b8ff: sets TABLASSERT_FULLMAP_SHARDS=2, builds via the public build_fullmap_db, and asserts META.shards="16" with all 16 sibling shard files present (and no s16) plus a 50-term lookup. The layout tests call build_fullmap_inner / build_test directly, so this closes the gap where a re-wired public path could go unnoticed. Env var is set then removed (house pattern, edition 2021), and nothing else reads it. Rust gate green: 56 unit + 10 golden, clippy -D warnings clean.
• shard_count_of normalization (major) — reasoned decline in the inline thread. Pre-existing, deliberate, documented read-path behavior unchanged by this PR (only comments were edited here). Rejecting malformed META.shards is a separate read-path design change; offered as a follow-up issue.