fix: preserve index entries when a parser fails instead of purging them

[GoodbyePlanet]

Summary

• registry.parse_file previously caught every parser exception and returned [], which the indexing pipeline treated as "file has no indexable symbols" and used as a signal to delete existing index entries for that file (pipeline.py).
• Because the file's blob SHA doesn't change when a parser merely crashes on it, incremental indexing would never retry the file — the deleted chunks were gone for good until the file's content changed.
• This introduces a dedicated ParseError exception: registry.parse_file now re-raises parser failures instead of swallowing them, and IndexPipeline.index_service catches ParseError to skip the file for this run while leaving its existing index entries untouched. Genuinely symbol-less files (parsed OK, zero symbols) still go through the existing delete-stale-entries path.

Fixes #68

Test plan

• uv run pytest tests/parser/test_registry.py — updated test_parse_file_swallows_parser_exceptions → test_parse_file_raises_parse_error_for_parser_exceptions, now asserts ParseError is raised
• uv run pytest tests/test_pipeline.py — new test_parser_failure_preserves_existing_index_entries asserts delete_by_file/upsert_chunks are not called when parsing raises
• uv run pytest — full suite green (217 passed)
• uvx ruff format --check on all changed files

🤖 Generated with Claude Code