chore(ast): remove dead call-tuple element + delete extract_calls_generic + close Swift/Rust test gaps (follow-on to #249, closed by #261)

[cdeust]

Follow-on to #249 — that issue's own resolver-dependent typecheck defect was independently closed by #253/#251 (merged as #261, now on main) via the same mechanism this PR originally proposed (a TypeGuard narrowing language: str to the language pack's own SupportedLanguage Literal, with AST_SUPPORTED derived from _EXTRACTORS rather than restated). Rebasing this branch onto current main surfaced that overlap directly in ast_parser.py; this PR now keeps main's mechanism and drops this branch's own duplicate _SupportedLanguage/_is_ast_supported/get_args(...) construction, which would otherwise have shipped a second, divergent narrowing beside the one already merged.

What this PR still does (the part #261 didn't)

Boy-scout pass (§14) on mcp_server/core/ast_parser.py / ast_extractor_registry.py, from the original #249 investigation:

• Dead code removed. Every per-language extractor (_extract_python/_extract_js/_extract_go/_extract_swift/_extract_rust, plus the JVM/C-family/scripting extractors ast_extractor_registry.build_extra_extractors composes) computed a flat calls list via extract_calls_generic and returned it as a 3rd tuple element that parse_file_ast immediately discarded — superseded by calls_per_function, the value actually consumed downstream. Extractor is now a 2-tuple (imports, definitions).
• extract_calls_generic itself deleted (review finding). Stripping that 3rd tuple element left extract_calls_generic (mcp_server/core/ast_extractors.py) with zero production callers — only its own two-test TestCallExtraction class called it directly. Deleted the function and that test class, corrected the module docstring's now-false "Also provides the generic call-site extractor used by all languages" claim, and reworded ast_extractor_registry._make_extractor's docstring to record that the function itself, not just its tuple slot, is gone. _MAX_CALL_NAME_LEN/_walk_type stay live — both still back extract_calls_per_function, the caller-qualified version parse_file_ast actually consumes.
• extract_calls_per_function's 40-line/method cap violation (boy-scout, seen while in this file for the deletion above). Pre-existing on main (51 lines including a nested walk closure), unrelated to either fix in this branch, but squarely in this diff's blast radius once the file was touched. Split into extract_calls_per_function (17 lines, contract unchanged) → _walk_for_calls (36 lines, the former closure, now a named helper taking the accumulator as an explicit parameter) → _collect_call_basenames (10 lines, the former inner per-call-type loop). Pure extract-method refactor, zero behavior change, verified by the unchanged test suite passing unmodified.
• Not addressed (reported, not filed): this file is still 330 lines, over the repo's 300-line/file cap (it was already 337 before either commit in this branch touched it — this PR only ever reduces it). A real fix needs a genuine module split; the file has 6 importers (ast_parser.py + the clike/extra/jvm/scripting extractor siblings + its own test file — Move-7's >5-importers High-stakes trigger), so splitting it means touching import statements in 5 other files. That's a separate, dedicated behavior-preserving-refactor PR per coding-standards.md §15.1, not a rider on a rebase-plus-review-finding PR that already carried real risk from two prior CONFLICTING states. Left for the coordinator to schedule rather than filed as an issue per this round's explicit no-new-issues instruction.
• Zero test coverage closed. _extract_swift/_extract_rust were never exercised through parse_file_ast by any existing test (the Go equivalent bypassed its own wrapper by calling extract_go_definitions directly). Added TestParseFileGo/TestParseFileSwift/TestParseFileRust.
• Smaller gaps closed: both branches of is_available(), _node_text/_extract_module_doc's docstring/comment-extraction branches (fake-Node unit tests), content_hash's length invariant, calls_per_function's populated content, and malformed-UTF-8 decode robustness (errors="replace").

Filed #269 for 50 mutmut "survived" reports left after the original mutation pass — independently verified as false positives (looping MUTANT_UNDER_TEST over every one against the correct, full test selection: every single one fails). Root cause: mutmut's coverage-based test-selection optimizer undercounts callers of a dispatch table (_EXTRACTORS) built once, eagerly, at module-import time — a tooling limitation orthogonal to this issue.

Rebase reconciliation (rebased onto main twice — after #261/#272, then again after #264/#268/#271/#274)

• ast_parser.py / ast_extractor_registry.py: resolved on the merits — kept main's _is_ast_language/pack-SupportedLanguage/derived-AST_SUPPORTED mechanism, kept this branch's 2-tuple dead-code removal on top of it (the two are independent and compose cleanly; ast_extractor_registry.py auto-merged with zero conflicts both times).
• .github/workflows/ci.yml: kept main's "Record the resolved type surface" step (prints pyright + tree-sitter + tree-sitter-language-pack versions), which is a superset of this branch's own single-package version-print step; dropped the latter as redundant.
• Suite-size files (CLAUDE.md, README.md, CONTRIBUTING.md, .bestpractices.json, docs/ASSURANCE-CASE.md, assets/badge-tests.svg): recomputed twice from a live pytest --collect-only -q on this machine each time — never a delta from either side's advertised number (this machine's collection is proven to match CI's canonical exactly, per fix(core): honour or refuse a stated timezone in normalize_date_to_iso (#252) #259). First rebase landed at 6573 (main's 6550 + this branch's +23); deleting extract_calls_generic's own two-test class dropped it to the final 6571. scripts/check_doc_claims.py --test-count 6571 and scripts/generate_repo_badges.py --check --test-count 6571 both green, badge regenerated by the actual generator script and diffed byte-identical against the manual conflict resolution.
• CHANGELOG's #249 entry rewritten: no longer claims to close typecheck gate is resolver-dependent: pyright reports 1 error under uv.lock, 0 under CI's pip resolution (ast_parser.py:88) #249 (already closed by fix(deps): pin tree-sitter-language-pack to the range where get_parser().parse(bytes) works (#253) #261); describes only the boy-scout dead-code-removal (including the extract_calls_generic deletion) and test-coverage work this branch still contributes. A stale "6549 → 6572" prose claim inside the entry itself (not part of any rebase conflict — git doesn't touch unconflicted prose) was caught by grepping the file content post-rebase, not just checking git status, and corrected to "6550 → 6571".

Completion Ledger

Path introduced by this diff	Asserting test
_extract_python/_extract_js/_extract_go/_extract_swift/_extract_rust 2-tuple return	TestParseFilePython, TestParseFileTypeScript, TestParseFileGo, TestParseFileSwift, TestParseFileRust (imports + definitions assertions)
ast_extractor_registry._make_extractor/build_extra_extractors 2-tuple return	TestParseFileTypeScript + test_ast_extractors_multilang.py (java/kotlin/c/cpp/csharp/ruby/php)
extract_calls_generic deletion — no remaining reference in mcp_server/ or tests_py/	grep -rn extract_calls_generic . (repo-wide, post-deletion) returns only the explanatory prose in ast_extractor_registry.py's docstring and this CHANGELOG entry, no code reference
extract_calls_per_function → _walk_for_calls + _collect_call_basenames split (behavior-preserving)	test_calls_per_function_populated + the full TestParseFile* family (all pass unmodified — same assertions, same call graph output)
is_available() True branch	test_is_available_reflects_installed_tree_sitter
is_available() False branch (ImportError)	test_is_available_false_when_tree_sitter_import_fails
_node_text/_extract_module_doc branches (7 distinct paths)	TestNodeText, TestExtractModuleDoc (7 focused fake-Node tests)
content_hash length invariant	test_content_hash_length
calls_per_function populated content	test_calls_per_function_populated
malformed-UTF-8 decode robustness (errors="replace")	test_decodes_malformed_utf8_without_raising, TestNodeText::test_replaces_invalid_utf8_instead_of_raising
main's _is_ast_language/SupportedLanguage-derived AST_SUPPORTED (unchanged by this PR, verified still intact post-rebase)	test_ast_parser_language_contract.py (all cases), test_ast_extractors_multilang.py::test_all_langs_registered
Recomputed suite count (6571) in doc-claim files	scripts/check_doc_claims.py --test-count 6571, scripts/generate_repo_badges.py --check --test-count 6571 (both run manually; also enforced live in CI's Lint/doc-claims jobs)

Test plan

• pytest tests_py/core/test_ast_parser.py tests_py/core/test_ast_extractors_multilang.py tests_py/core/test_ast_extractors.py tests_py/core/test_ast_parser_language_contract.py: 69 passed
• pytest tests_py/core/: 3384 passed (no regression)
• pytest (full suite): 6571 passed, 117 subtests passed — exact match to pytest --collect-only -q, run twice (before and after the extract_calls_per_function split)
• ruff check + ruff format --check on all changed .py files: clean
• pyright on changed .py files: 0 errors, 0 warnings
• scripts/check_doc_claims.py --test-count 6571 / scripts/generate_repo_badges.py --check --test-count 6571: both green
• tests_py/scripts/test_typecheck_env_parity.py, test_check_doc_claims.py, test_generate_repo_badges.py: 77 passed, 30 subtests passed
• CI on pushed head b8fb187: all 19 checks green (gh pr checks 270)

🤖 Generated with Claude Code

https://claude.ai/code/session_012Yu6EnWspTfqHoGkExyS6u