feat: add flake-aware Nix language support

[BrockLeeBytes]

Summary

Adds .nix as a supported language in the Tree-sitter parser with flake-aware semantics. No new runtime dependency — tree-sitter-language-pack already ships the Nix grammar.

What gets extracted:

• Top-level and nested attrset bindings (attrpath = expr;) become Function nodes with flattened dotted names (e.g. packages.default, devShells.default). CONTAINS edges wire the file to each top-level binding.
• In flake.nix, inputs.<name>.url = "..." strings (both the flat and the nested inputs = { nixpkgs.url = "..."; ... }; forms) emit IMPORTS_FROM edges targeting the URL.
• import <path> and callPackage <path> <args> applications in any .nix file emit IMPORTS_FROM edges. Relative paths are resolved against the caller's directory (mirrors the Bash source handling); URLs and bare identifiers are preserved as-is.

Changes

• code_review_graph/parser.py:
  • EXTENSION_TO_LANGUAGE: .nix → nix
  • Empty entries in _CLASS_TYPES / _FUNCTION_TYPES / _IMPORT_TYPES / _CALL_TYPES (pattern matches Elixir/Bash)
  • New dispatch in the main walker for binding nodes
  • New _extract_nix_constructs method + three helpers (_is_nix_flake_file, _nix_attrpath_parts, _extract_nix_flake_input_urls, _extract_nix_import_targets)
  • _do_resolve_module: new nix branch for relative .nix path resolution
• tests/fixtures/sample.nix, tests/fixtures/sample_module.nix: flake and non-flake fixtures
• tests/test_multilang.py: new TestNixParsing class (7 tests covering language detection, node language tags, top-level binding flattening, flake inputs.*.url edges, import/callPackage edges, non-flake negative case, and CONTAINS wiring)
• CLAUDE.md, README.md, CHANGELOG.md, diagrams/generate_diagrams.py: docs and language-coverage diagram updated

Verification

Automated:

uv run pytest tests/test_multilang.py::TestNixParsing -v    # 7 passed
uv run ruff check code_review_graph/parser.py               # clean
uv run mypy code_review_graph/parser.py --ignore-missing-imports --no-strict-optional  # clean

Full suite: +7 new passes, 0 regressions (the 60 pre-existing failures in TestJuliaParsing / TestGDScriptParsing / test_main's TestApplyToolFilter are all unrelated to this change — confirmed by running the full suite on HEAD vs. HEAD~3).

Test plan

• uv run pytest tests/test_multilang.py::TestNixParsing -v — all 7 pass
• uv run pytest tests/ --tb=short -q — no new regressions
• uv run ruff check code_review_graph/parser.py — clean
• uv run mypy code_review_graph/parser.py --ignore-missing-imports --no-strict-optional — clean
• E2E build on a real multi-file flake — parses, produces expected nodes/edges via MCP tools

Notes

• The non-flake test (test_non_flake_file_has_no_input_edges) verifies that the inputs.*.url branch only fires when the file basename is flake.nix, so random .url bindings in NixOS modules don't get misinterpreted as flake inputs.
• The flake-input branch on the nested form (inputs = { nixpkgs.url = "..."; ... };) is necessary — the flat form (inputs.nixpkgs.url = "...";) only covers part of real-world flake usage.