compat: wire the 42-module parity matrix into CI with per-module trend

[proggeramlug]

Part of #793 (axis 3 — Node.js platform built-in modules with real semantics).

What already exists (committed, not yet CI-wired)

A behavioral parity suite was built and committed over the last week:

• test-files/test_parity_<module>.ts — 42 files, one per supported node:* module. Every API in the parity reference becomes a console.log("label:", expr) line under per-section try/catch. The byte-for-byte diff against node --experimental-strip-types IS the per-module gap report. Commits: d0e84724, 08c9004a, 1f2f44e5.
• docs/runtime-parity.md — leaf-level inventory of the entire Node + Bun stdlib API surface (~50 modules, ~4,200 rows: | API | Node.js | Bun | Notes |), sourced from the authoritative Node docs + Bun compat table.
• docs/runtime-parity-gaps.md — that inventory diffed against Perry's compile-time coverage (manifest + Expr::* HIR variants + js_* FFI exports across perry-runtime / perry-stdlib / 35 perry-ext-* crates). 2,417 true gaps.
• scripts/gen_parity_tests.py + scripts/parity-skiplist.toml — regenerate skeletons from the inventory; skiplist documents architecturally-out-of-scope modules (vm/inspector/v8/etc.) with rationale.
• run_parity_tests.sh --filter parity_ — runs just the parity matrix; auto-PERRY_ALLOW_UNIMPLEMENTED=1 so unimplemented APIs surface as runtime divergence instead of hard compile errors; strips DEP/Experimental Node warnings from the diff.

This is ~70% of the axis-3 instrumentation #793 is asking for. It is currently run ad-hoc locally and has repeatedly been cleaned out of the working tree between sessions (now committed, so durable).

What this sub-issue adds

Wire it into CI as a tracked, per-module signal — distinct from #794 (per-category thresholds) and #800 (Node's own test corpus):

1. CI job runs ./run_parity_tests.sh --filter parity_ on every push to main.
2. Per-module trend artifact — emit a table of module | node_lines | perry_lines | diff_lines | status so a regression in any single module is visible (not buried in a global %). The diff-line count per module is already computed locally; just persist it as a CI artifact + job annotation.
3. Baseline lock — store the current per-module diff counts as test-parity/parity_matrix_baseline.json; CI fails if any module's diff grows (regression) and prints a celebratory note when one reaches 0 (full byte-for-byte PASS).
4. Hard per-test output cap — see the robustness note on CI: stop truncating gap-suite output, add per-test summary #796: one test (test_parity_timers_promises, root cause await on namespace-member that returns undefined re-enters catch arm infinitely (5.7M lines observed) #712) emitted 5.7M identical lines and the runner's O(n²) bash normalize_output burned ~3 hours. The CI job MUST cap per-test stdout (e.g. 50k lines) and the normalizer must be linear, or pathological output DOSes the run.

Current baseline (v0.5.910 sweep, for reference)

• PASS (diff = 0): os, tty, async_local_storage (+ argon2, ethers as expected-output).
• One fix from PASS: path (diff = 4, blocked only on path.posix / path.win32 sub-namespace methods return undefined (last gap before path PASS) #810 — path.posix/win32 sub-namespace).
• Near-pass (diff < 35): string_decoder, async_hooks, tls, readline.
• 18 compile-fails: modules with import * as X from "node:X" where Perry has no stdlib backing — strict-import policy (Namespace import for unrecognized node:* sub-paths resolves to true (TAG_TRUE) #629 / v0.5.835) correctly rejects them; these need real native bindings, not test changes.
• Large surface gaps: http (307), fs (207), stream (141), zlib (130), net (115), buffer (93).

Trend across sweeps v0.5.713 → v0.5.910: 0 → 5 PASS as #623/#629/#630/#631/#648/#649/#650/#681/#682/#712/#741/#742 landed.

Done when

• A CI job produces the per-module diff table as an annotation on every main push.
• parity_matrix_baseline.json is committed and regressions fail CI.
• Per-test output is hard-capped and normalize_output is linear-time.

[andrewtdiz]

I am taking a focused CI/reporting slice for the Node parity matrix.

Planned PR scope:

• add a per-module trend/baseline checker that reads test-parity/reports/latest.json plus test-parity/known_failures.json
• emit a stable module table and JSON artifact with pass/fail/compile-fail counts
• fail only on new untriaged failures or regressions against a committed baseline
• wire the artifact into the existing parity CI job without changing runtime module behavior

Non-goals for this cut: importing upstream Node core tests (#800), changing module implementations, and editing current module-inventory known-failure rows.

[andrewtdiz]

Opened draft PR #4286 for the focused parity-matrix trend/reporting gate: #4286

Scope covered:

• per-test result records in the parity runner report
• scripts/parity_matrix_trend.py to emit/check module-level status, known-failure state, output line counts, and diff-line counts
• committed baseline entries for the current module-inventory rows
• CI step-summary and artifact wiring for the matrix report

I left upstream Node core corpus import (#800), runtime module fixes, and known_failures.json cleanup out of this cut.

[andrewtdiz]

I am taking a narrow stale module-inventory known-failure cleanup slice that complements, but does not overlap, draft PR #4286.

Scope for this PR: remove only the test_parity_zlib row from test-parity/known_failures.json after validating that both the generated test_parity_zlib inventory probe and the curated node-suite/zlib suite pass on current origin/main with Node 26.

Non-goals: parity-matrix CI/trend-gate wiring from #4286, HTTP/DNS/SQLite/stream/TLS inventory rows, and any runtime zlib behavior changes.

[andrewtdiz]

Opened draft PR #4296 for the narrow zlib inventory skip-list cleanup: #4296

Scope:

• removes only the stale test_parity_zlib entry from test-parity/known_failures.json.
• leaves the parity-matrix trend gate ([codex] add parity matrix trend gate #4286) and other module-inventory rows untouched.

Validation on current origin/main with Node 26:

• test_parity_zlib: 1 pass / 0 fail / 0 compile fail
• full node-suite/zlib: 57 pass / 0 fail / 0 compile fail
• jq empty test-parity/known_failures.json
• cargo fmt --all -- --check
• git diff --check
• ./scripts/check_file_size.sh