Avoid quadratic C header evidence scans

[Ha1baraA11]

Linked issue

Fixes #977

What & why

• The C-header evidence scan called _is_in_static_dead_guard() for every syntax
  node. On long nested #elif chains, each call rescanned ancestors. This PR
  carries dead-branch state through one top-down scan and makes other guard
  checks inspect each parent-child edge once, preserving if false and #if 0
  behavior.
• On the synthetic 800-branch header, direct CodeParser.parse_bytes time fell
  from 108.417 seconds (baseline) to 0.231 seconds (patch) on macOS 15.8
  arm64/Python 3.13.9. This timing does not cover a full CLI build.

How it was tested

• uv run pytest tests/ --tb=short -q — 4269 passed, 850 skipped, 2 xfailed,
  2 xpassed; one Starlette deprecation warning. A temporary 20 ms
  PollingObserver shim avoided an unrelated macOS FSEvents crash; no project
  file was changed for the shim.
• uv run ruff check code_review_graph/ — passed.
• uv run ruff check code_review_graph/ tests/test_cpp_qt_headers.py — passed.
• uv run pytest tests/test_cpp_qt_headers.py --tb=short -q — 15 passed.
• uv run mypy code_review_graph/ --ignore-missing-imports --no-strict-optional
  reported one type error at code_review_graph/embeddings.py:253, outside this
  PR's changed files.

Checklist

• Tests added for new functionality
• All tests pass: uv run pytest tests/ --tb=short -q
• Linting passes: uv run ruff check code_review_graph/
• Type checking passes:
  uv run mypy code_review_graph/ --ignore-missing-imports --no-strict-optional
• Lines are at most 100 characters
• Docs updated where behavior changed (README, docs/, docstrings)