test: reduce registry manifest test repetition

[PascalThuet]

What

Reduce the expensive multi-install manifest contract from one CLI setup per safe-integration pair to two aggregate install orders: forward and reverse.

This keeps the pairwise manifest-disjoint assertion, but builds the manifests once per order instead of repeating the same setup for every pair.

Stack

Stacked on #3144. Because this fork cannot push the stack base branch to github/spec-kit (403), this draft PR targets main for upstream visibility and will include the #3144 diff until #3144 lands. After #3144 is merged, this PR diff should shrink to only tests/integrations/test_registry.py.

Validation

• uv sync --extra test
• uvx ruff check tests/integrations/test_registry.py
• git diff --check
• .venv/bin/python -m pytest tests/integrations/test_registry.py --durations=30 -q (1035 passed in 34.86s)
• .venv/bin/python -m pytest tests/integrations --durations=30 -q before the final naming/comment cleanup (2500 passed, 1 skipped in 236.69s)

Comparison on the same machine:

• Base stack PR: tests/integrations/test_registry.py collected 1223 tests and took 248.55s
• This PR: tests/integrations/test_registry.py collects 1035 tests and took 34.86s

Disclosure

Prepared by Codex (model: GPT-5) on behalf of @PascalThuet.

[PascalThuet]

Pushed ca07e0e for the review feedback on test_safe_integrations_have_disjoint_manifests. The assertion is unchanged; this is clarity plus a guard.

• Documented why both orders run. Manifest contents don't depend on install order: each safe integration writes only to its own disjoint directories and always records what it writes, so installing the whole set into one project gives the same per-integration manifests either way. forward/reverse doesn't vary the manifests; it just routes a different integration through init vs integration install (forward inits the first key, reverse the last). The old comment didn't say this and read as if order mattered.
• Added a len(ordered_keys) >= 2 precondition. With fewer than two safe integrations the contract is vacuous and ordered_keys[0] would fail on an empty set, so a shrunken registry now fails here instead of passing silently.
• Built the manifest map with a comprehension and pulled out integrations_dir. Cosmetic.

Validation: ruff check clean; pytest tests/integrations/test_registry.py -> 1035 passed in 6.48s.

(Disclosure: drafted and verified with an AI coding agent; reviewed by me.)

[Copilot]

Pull request overview

This PR refactors the multi-install-safe integration manifest contract test to reduce runtime by installing all safe integrations only twice (forward and reverse order) instead of repeating full CLI setup for every safe-pair combination. As part of the stacked changes, it also isolates integration-test HOME/XDG directories to prevent tests from reading/writing the real user home.

Changes:

• Add an autouse pytest fixture to redirect HOME/USERPROFILE and XDG directories into each test’s tmp_path.
• Replace the pairwise “init + install” manifest-disjointness loop with two aggregate install orders (forward + reverse) and then assert pairwise manifest disjointness from the resulting manifests.

Show a summary per file

File	Description
tests/integrations/test_registry.py	Refactors the manifest disjointness test to use two aggregate install orders and then check all safe-pairs for overlap.
tests/integrations/conftest.py	Adds an autouse fixture to isolate HOME/XDG paths for integration tests.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

• Files reviewed: 2/2 changed files
• Comments generated: 1

[mnriem]

Please address Copilot feedback. If not applicable, please explain why