fix(insight): handle individual LLM failures in qualitative insights (#2341)

[netbrah]

Fixes #2341

Root Cause

generateQualitativeInsights runs 8 parallel LLM calls via Promise.all. The inner generate helper re-throws errors, so a single LLM call failure (timeout, rate limit, JSON parse error) rejects the entire Promise.all, catches at the outer try/catch, and returns undefined for the whole qualitative object.

In the React report, every detailed section is gated by {data.qualitative && (...)} — so when qualitative is undefined, all sections disappear. Only the header and stats row survive.

Fix

• DataProcessor.ts: Changed the generate helper to catch errors and return undefined instead of re-throwing. Individual LLM call failures no longer crash the entire qualitative section — surviving calls still populate their sections.
• QualitativeInsightTypes.ts: Made all fields on QualitativeInsights optional so partial results are type-safe.
• Each React section component already handles missing data (if (!field) return null), so no frontend changes needed.

Test

Added 3 unit tests:

• Partial failure: some LLM calls fail → qualitative is defined, failed sections are undefined, successful sections populate
• Empty facets: returns undefined as before
• Full success: all 8 calls succeed, qualitative is complete

Made with Cursor

[DragonnZhang]

LGTM!