perf(diff-viewer): optimize buildFileEntries comment grouping

[matt2e]

Summary

• Eliminates O(files × comments) nested loop in buildFileEntries by pre-grouping comments into a Map keyed by basename (filename)
• Each file now only checks pathsMatch against the small set of comment paths sharing its basename, rather than scanning all comments
• With 200 files and 50 comments, this reduces pathsMatch calls from ~10,000 to a handful per derived state update

Details

The buildFileEntries function is called from a $derived in DiffModal.svelte, so it re-runs on every state change (file review toggling, new comments, etc.). The previous implementation iterated all comments for every file. The new implementation:

1. Groups comments by basename into a Map<basename, Map<commentPath, {count, types}>> in a single O(comments) pass
2. For each file, looks up candidates by basename (O(1)) and only runs pathsMatch against that small group
3. Preserves the existing pathsMatch suffix-matching semantics exactly

Test plan

• All existing pathsMatch tests pass (10/10)
• All staged app tests pass (28/28)
• svelte-check passes with 0 errors, 0 warnings
• Open a diff modal with a PR that has many files and many comments; verify comment counts still show correctly on the file tree sidebar
• Toggle file review status and add comments; verify the diff modal remains responsive

🤖 Generated with Claude Code