refactor(standards): split language_standards.py into a per-language package

[squid-protocol]

Summary

gitgalaxy/standards/language_standards.py had grown to 14,403 lines / 942KB across 59
languages -- the largest file in the repo, and the one place that hadn't adopted the
per-language-file convention already used by tests/extraction/languages/ and
docs/language_status/. Every consumer (detector.py, language_lens.py, prism.py,
galaxyscope.py, the supply-chain tools, ~150+ tests) only ever accesses
LANGUAGE_DEFINITIONS as an already-merged, in-memory dict via constructor injection or
whole-dict import -- nothing depends on the file being physically monolithic, which is what
makes this safe as a pure structural refactor.

gitgalaxy/standards/language_standards.py is now a package:
gitgalaxy/standards/language_standards/__init__.py assembles the exact same merged dict
(same 59 keys, same order) from one file per language under languages/, plus sibling modules
for the genuinely cross-language pieces (_lens_config.py, _prism_config.py,
_shared_patterns.py's GLOBAL_* debt/AI-SDK detectors, _overrides.py's
PROJECT_OVERRIDES). Every existing import path
(from gitgalaxy.standards.language_standards import LANGUAGE_DEFINITIONS, etc.) is
unchanged, so no consumer -- including language_lens.py (the language-identity classifier)
-- needed any code change.

The split was done as a source-text slice (ast.get_source_segment per top-level dict entry
against the git-committed source), not a load-then-repr() round-trip, specifically to
preserve every in-block # comment (real documentation throughout this registry) and avoid
any risk of altering a regex string literal's escaping.

Companion changes in this PR (all mechanically required, not optional):

• gitgalaxy_config.py's XRAY_BYPASS_PATHS exact-file entry -> a directory prefix (the
  binary-anomaly-detector's bypass check is a substring match, so this still covers every new
  file underneath).
• tree_sitter_accuracy_audit.py's _LANGUAGE_STANDARDS_PATH repointed to the new
  __init__.py (where the auto-generated accuracy summary table now lives, byte-identical).
• Four GitHub workflow files' paths: triggers and exact-file diff/commit checks updated
  (tree-sitter-accuracy-audit.yml/-history.yml, tri-comparison-audit.yml/-history.yml)
  since they gate on or auto-commit to this exact path.
• how_to_add_a_language.md's registration steps, the standards README.md's module
  breakdown, and the harden-language-extraction skill + class-start-scout/
  strict-signature-scout subagents updated to describe the new per-language file layout
  (a handful of purely-prose mentions elsewhere are left for a follow-up sweep, since the
  dotted import path itself didn't change).

Type of change

• Bug fix
• New feature / language support
• Parsing or engine logic (gitgalaxy/core/detector.py, language_standards.py, prism.py, or a per-language rule)
• Docs, tooling, or CI only
• Other (describe above)

CI checklist

• python tests/tools/audit_check.py passes (bundles the ruff/mypy/dead-key baseline-gated audits + ruff format --check)
• python -m pytest tests/ passes
• python tests/tools/crucible_check.py -- see Verification below, drift is pre-existing/environmental, not caused by this PR
• python tests/tools/tree_sitter_accuracy_audit.py --all --ci -- see Verification below, same environmental caveat for 4 of 35 checked languages

Differential Scan target

Not applicable -- pure structural refactor, no regex/parsing behavior changed (see the
byte-for-byte equivalence check below).

Verification

$ python /tmp/verify_split_equivalence.py   -- deep-compares every language's rules dict
  (patterns + flags) plus LENS_CONFIG/PRISM_CONFIG/PROJECT_OVERRIDES/
  HTML_NONEXECUTABLE_SCRIPT_TAG against `git show origin/main:...language_standards.py`
  OK: 59 languages, all byte-for-byte equivalent, key order preserved

$ python /tmp/verify_comments.py            -- every in-block comment survived the split
  OK: every language's comment set matches exactly

$ ruff format --check .                     -- clean
$ python tests/tools/audit_check.py         -- [ruff] OK, [mypy] OK, [dead-key] OK, [ast-accuracy] OK
$ python -m pytest tests/                   -- 7138 passed, 2 skipped, 9 xfailed, 3 xpassed, 0 test files touched
$ python tests/tools/tree_sitter_accuracy_audit.py --summary-table   -- zero diff (confirms the repointed path)
$ python tests/tools/tri_comparison_chart.py --all --ci              -- all OK (3 languages with committed precision baselines)
$ fresh `pip install -e ".[yaml]"` + import sanity check              -- 59 languages load correctly
  (catches a missing package __init__.py that a stale editable install would mask)
$ galaxyscope self-scan of the new package --fail-on-secrets          -- zero secrets-risk findings,
  confirming the split didn't need to propagate the galaxyscope:ignore marker beyond __init__.py

crucible_check.py and tree_sitter_accuracy_audit.py --all --ci both show drift in this
environment (4061/4054 golden-master diffs; 4 of 35 languages failing a "ground-truth corpus
changed" check). Both were verified as pre-existing and unrelated to this PR by stashing all
changes and re-running against pristine, unmodified origin/main: the golden-master diff count
matched exactly (4061/4054), and the same 4 languages failed with the identical
"files_scanned: baseline=X current=Y ... This means the corpus changed" message (a stale local
language-crucible checkout, not a regex regression -- the tool itself distinguishes this
failure mode from an actual accuracy regression). Combined with the byte-for-byte equivalence
check above, this confirms zero behavioral change from the split.

Worktree rebase note: 8 active local worktrees carry small, isolated single-language diffs
to the old language_standards.py (fix-python-2238, fix-csharp-2237, fix-ruby-2241,
fix-jcl-2482, fix-cpp-2010, fix-cpp-2009, fix-cpp-2012, fix-perl-2239) and will hit a
real delete/modify conflict rebasing onto this -- expected and loud, not silent; each one maps
cleanly onto its language's new file under languages/.

Co-Authored-By: Claude Sonnet 5 noreply@anthropic.com

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab