fix(parser): resolve custom-language names via configurable name_field (#691)

[anusbutt]

Linked issue

Closes #691

What & why

The generic walker's name heuristic only checked an identifier-like child or a
field literally named name, so common BibTeX/LaTeX/Markdown node shapes were
extracted unnamed and silently dropped (losing the node and all its edges).

Adds an optional name_field key to .code-review-graph/languages.toml
(string or ordered list). For custom languages only, each candidate is resolved
field-first across all candidates, then by descendant node type, then descended
to a cleaned text leaf. Two-pass ordering makes precise fields win over a
broader type match (e.g. LaTeX \newcommand resolves via declaration, not the
text node in its body). Anchoring on configured candidates avoids returning
unrelated inner identifiers (e.g. BibTeX field labels title/author).
Built-in language extraction is byte-for-byte unchanged.

How it was tested

uv run pytest tests/test_custom_languages.py -q   # 31 passed
uv run pytest tests/test_parser.py tests/test_multilang.py -q  # 516 passed, no regressions
uv run ruff check code_review_graph/              # clean
uv run mypy code_review_graph/ --ignore-missing-imports --no-strict-optional  # clean
Full suite: 1972 passed / 5 skipped / 2 xpassed. One unrelated failure
(test_mcp_stdio_shutdown) is pre-existing — it fails identically on a clean
tree (a 20s MCP-server-startup timeout), not touched by this change.

Checklist

- [x] Tests added for new functionality
- [x] All tests pass (except one pre-existing, unrelated MCP-stdio-shutdown timeout)
- [x] Linting passes
- [x] Type checking passes
- [x] Lines are at most 100 characters
- [x] Docs updated (docs/CUSTOM_LANGUAGES.md)

🤖 Generated with Claude Code (https://claude.com/claude-code)