perf(metrics): retire the remaining Node::parent climbs (#1088)

[dekobon]

Closes #1088.

Retires the four Node::parent climbs #1084 deferred — and one cause
the issue did not name, which turned out to be the one that mattered.

What changed

perf(node): scan a node's children with a cursor — Node::wraps_any,
reached from is_child and has_sibling and so from most languages'
checkers and getters, walked a node's children with child(0) +
next_sibling(). #217 chose that form over the cursor iterator to avoid
a TreeCursor allocation, on the stated premise that a sibling step is
O(1). It is not: ts_node_next_sibling resolves the parent first, and
tree_sitter stores no parent pointer — it descends from the root. The
scan was O(children × depth), making every caller O(depth) per node.

perf(metrics): thread the chain through the last climbs — three of
the four listed sites take an Ancestors parameter (JS-family
is_func / is_closure and the JS space-name getters, Elixir's
get_func_space_name, the suppression scan), plus Ruby's is_closure,
which the first pass missed. ops_inner, act_on_node, and
suppression_markers maintain chains of their own. All widened items
are pub(crate); the published API is unchanged.

The fourth site wanted deleting, not threading. Elixir's Npm / Npa
compute opened with is_func_space_with_code, whose answer the
defmodule keyword check immediately below always discarded —
elixir_is_class_macro is exactly defmodule, so the classifier
admits precisely what that check admits, and every node whose
classification consulted the quote-template ancestor walk is rejected
by it regardless. Removing the call removes the work rather than making
it cheaper, and leaves both signatures untouched.

Measurements

The issue's "worth measuring first" instinct is what caught the real
cause: threading the chain into the JS predicates left the probe at
k = 1.97. The walk is two steps long on that shape, so it should have
flattened — the remaining O(depth) was one level down, in
wraps_any. A chain-fed predicate is only as linear as the
primitives it calls.

Probe	before	after
nom/nested-arrow (JS, one arrow per level)	k = 1.97, 17.6 s @ depth 4000	k = 1.03, 6.3 ms
nom/nested-declared-function (shape control)	k = 1.08	k = 1.08

Ordinary input benefits too: a metric walk over the 384-file pdf.js
corpus drops from ~443 ms to ~370 ms.

The probe deliberately uses block-bodied arrows, where the walk stops
in two steps. A chain of expression-bodied arrows
(a => b => c => …) is the one shape no chain can flatten — no stop
kind intervenes, so the walk itself is O(depth) long. Documented at
the macro rather than silently left.

Metric values

Unchanged for every language. Full workspace suite green with no
.snap.new under the big-code-analysis-output submodule, so no
submodule bump is needed.

Tests

Chain-vs-climb parity for every newly threaded consumer, each verified
by perturbation to fail against a wrong ancestor read: is_func /
is_closure across all four JS grammars plus Ruby, get_func_space_name
across all four JS grammars plus Elixir, has_sibling, a JS fixture
added to the existing a_known_chain_answers_exactly_what_climbing_answers,
the bca function span walk, and act_on_node on a subtree.
wraps_any's equivalence test keeps its node-by-node comparison with the
reference flipped — the retired sibling chain is now the reference, so
what it pins is that the swap did not change the set of children
enumerated.

Two Npm / Npa parity tests written during the first pass were
vacuous and were replaced: they asserted that a known chain and a
climb agree, over a fixture where the ancestor answer never reached the
output — which is the redundancy described above, showing up as a test
that could not fail. The replacements pin counts over a defmodule
nested inside a quote.

Follow-up

#1096 tracks the remaining per-node node.parent() calls (Halstead
get_op_type getters, src/metrics/abc/, the loc / npm / npa /
cyclomatic bodies). docs/development/benchmarking.md is updated to
point there, and the Ancestors::unknown() call sites that remain are
deliberate and documented at each.

Validation

make pre-commit green, including the self-scan gate at both tiers.
.bca-baseline.toml net shrank — removing the redundant Elixir
classifier and reverting the unused Npm / Npa parameter took several
functions back below their recorded values.

[codecov]

Codecov Report

❌ Patch coverage is 97.71689% with 10 lines in your changes missing coverage. Please review.
✅ Project coverage is 98.08%. Comparing base (0e630cd) to head (1d6ff63).

Files with missing lines	Patch %	Lines
src/checker.rs	92.15%	4 Missing ⚠️
src/getter.rs	96.22%	2 Missing ⚠️
src/node.rs	97.40%	2 Missing ⚠️
src/suppression.rs	93.75%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1097      +/-   ##
==========================================
- Coverage   98.10%   98.08%   -0.03%     
==========================================
  Files         272      272              
  Lines       67527    67776     +249     
  Branches    67097    67346     +249     
==========================================
+ Hits        66249    66477     +228     
- Misses        868      888      +20     
- Partials      410      411       +1     

Flag	Coverage Δ
rust	98.07% <97.71%> (-0.03%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/checker/bash.rs	58.82% <100.00%> (ø)
src/checker/c.rs	76.92% <100.00%> (ø)
src/checker/cpp.rs	77.77% <100.00%> (ø)
src/checker/csharp.rs	90.90% <100.00%> (ø)
src/checker/elixir.rs	62.06% <100.00%> (-25.87%)	⬇️
src/checker/go.rs	76.92% <100.00%> (ø)
src/checker/groovy.rs	100.00% <100.00%> (ø)
src/checker/irules.rs	70.00% <100.00%> (ø)
src/checker/java.rs	86.95% <100.00%> (ø)
src/checker/kotlin.rs	69.56% <100.00%> (ø)
... and 38 more

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.