[codex] Parallelize incremental parse step - #1052

Summary

• Reuse the changed file parse result from update_file_in_graph() instead of parsing the same file again during incremental relinking.
• Parse remaining affected files with a bounded ThreadPoolExecutor driven by PARALLEL_WORKERS.
• Keep graph writes and relationship relinking serialized to avoid backend lock contention.
• Add regression coverage for the worker-count config and changed-file parse reuse.

Why

Incremental indexing was effectively serial for affected-file parsing, and the changed file could be parsed twice in one update pass. This makes large affected sets slower than necessary even when parsing work is CPU-bound and independent per file.

Validation

• PYTHONPATH=src /tmp/cgc-pr-venv/bin/python -m pytest tests/unit/tools/test_graph_builder_perf_fixes.py::TestWatcherIncrementalHandleModification::test_incremental_parse_workers_uses_parallel_workers_config tests/unit/tools/test_graph_builder_perf_fixes.py::TestWatcherIncrementalHandleModification::test_handle_modification_reuses_changed_file_parse_result -q
• PYTHONPATH=src /tmp/cgc-pr-venv/bin/python -m compileall -q src/codegraphcontext/core/watcher.py

Note: the default Homebrew Python in this shell is externally managed and lacked project dependencies, so validation used a temporary venv under /tmp/cgc-pr-venv.