fix(cpp): recover callables after preprocessor-gated signatures

[blankanswer]

Summary

Fixes #946.

C++ files that gate alternate out-of-line method signatures with #ifdef / #else / #endif can confuse the raw tree-sitter parse and cause methods after the conditional block to disappear from the graph.

This PR extends the existing C/C++/CUDA preprocessor second pass so it can recover missing Function / Method definitions from the expanded AST, while keeping the normal raw-source extraction as the primary source of truth.

What Changed

• Added a regression test for the SurfaceFlinger-style conditional signature shape.
• Added a lightweight definition-walker entry point that does not emit a Module node or variables.
• During the existing preprocessor second pass, merge only missing callable definitions from the expanded AST.
• Deduplicate recovered callables by label + qualified_name.
• Did not change vendored tree-sitter grammars.

Validation

• make -f Makefile.cbm SANITIZE= build/c/test-runner build/c/codebase-memory-mcp
• build/c/test-runner extraction — 211 passed
• build/c/test-runner c_lsp — 750 passed
• git diff --check
• CLI smoke test with a temporary C++ repo:
  • index_repository
  • search_graph verified both commit and composite are indexed after the #ifdef/#else/#endif signature block.

[DeusData]

Thanks for the focused C++ parser fix. Triage: parser recovery bug for #946, normal priority. Review focus is ownership of recovered callables, avoiding duplicate definitions across #ifdef branches, and keeping unrelated extraction behavior unchanged.

[DeusData]

This is the fix we most wanted from the 0.9.1-rc bug queue — the coverage signal we shipped this week (#963/#968) flagged exactly this class on our own codebase (#961: cbm_path_within_root, handle_process_kill — both in-function #ifdef splits), and recovering missing callables from the already-existing preprocessor second pass is the right systemic answer: raw parse stays primary, expanded AST only fills verified gaps, deduped by label+QN. The defs-only walker entry point is a clean seam.

Two things before merge:

1. Line coordinates of recovered defs. The merged callables carry EXPANDED-source line numbers, but get_code_snippet (and the UI's GitHub deep-links) read the ORIGINAL file at start_line..end_line — a drifted range shows the wrong code for exactly the methods this PR rescues. Please either remap to original coordinates where feasible, or mark recovered defs (e.g. a recovered_from_preprocessed property / reduced confidence) and document the drift — plus one test pinning whatever behavior you choose, so snippet consumers aren't silently wrong.

2. Rebase onto current main — the end-of-extraction region of internal/cbm/cbm.c changed in feat(index): best-effort parse-coverage signal — flag not-fully-indexed files #968: the parse-coverage signal now computes error regions AFTER extraction and subtracts recovered definitions. Your merge hook lands in the same neighborhood (conflict likely), and there's a pleasant interaction to verify once rebased: callables you recover should automatically UN-flag the corresponding parse_partial ranges (a region fully covered by recovered defs is dropped). The regression test tests/test_parse_coverage.c shows the mechanics; a one-line assertion in your test that the fixture is no longer parse_partial-flagged would prove the two features compose.

Genuinely excited about this one — with the rebase + the line-coordinate decision it merges.