perf: query timeout on large repos (80K+ files)

[avifenesh]

Problem

repo-intel query onboard and repo-intel query can-i-help time out (>30s) on very large repos like TypeScript (81K files) and Deno (28K files). The areas() function iterates all file_activity entries and groups by directory, which becomes expensive at scale.

Repos affected

• microsoft/TypeScript (~81K files)
• denoland/deno (~28K files)

Potential fixes

1. Cache areas() result within a query session (it's called by both onboard and can-i-help)
2. Pre-compute directory groupings during merge_delta() instead of on-the-fly
3. Add a file count threshold - for repos > 10K files, sample or limit to recent files only

Context

Discovered during 100-repo validation. 97/100 repos pass; these 2 timeout.