Optimize ingestion load testing and add multi-document chat querying

[varun3011]

Adds a follow-up optimization pass for ingestion load testing, workspace cleanup, and multi-document chat querying.

Ingestion and Load Testing

• Set default ingestion worker balance to 4 extract workers and 4 index workers.
• Add parallel signed-URL uploads to the ingestion load-test script with configurable upload concurrency.
• Optimize index worker DB writes by inserting embedding rows in one batch per document.
• Keep embedding token usage unchanged while reducing DB round-trips.
• Optimize ingestion-run status refresh by combining run lookup and status counting into one query.
• Add DB pool configuration to reduce Supabase connection pressure during worker load.
• Replace schema helper inspection with information_schema queries to avoid extra DB connections.
• Add a Delete all documents control in the Upload / Ingestion workspace page for faster test cleanup.

Multi-Document Chat Querying

• Add multi-document selection in the document sidebar.
• Add chat query modes:
  • Selected documents
  • All ready documents
• Enforce current query caps in the UI:
  • 10 documents max
  • 100 total pages max
• Send document_ids to the streaming query API.
• Keep existing document_id compatibility by using the first selected document as the primary document.
• Persist and restore selected document sets in saved chat sessions.
• Add chat_sessions.document_ids schema support.
• Improve multi-document retrieval by selecting balanced candidates per document before final ranking.
• Increase server DB pool defaults for streaming query reliability while keeping worker DB pools conservative.

Validation

• Verified Docker runtime with 4 extract workers and 4 index workers.
• Ran 20-document valid ingestion load tests.
• Best observed 20-document run:
  • Avg extraction: ~1.1s
  • Avg indexing: ~3.8s
  • Avg total ingestion: ~12s
  • Batch wall time: ~50s
• Manually tested querying across 3 uploaded/indexed PDFs.
• Confirmed selected-document mode works.
• Confirmed streaming answers and citations work.
• Confirmed DB pool timeout error is fixed.
• Worker regression tests passed: 2 passed.
• Focused server tests passed: 10 passed.
• Multi-document query/retrieval tests passed: 5 passed.
• Client build passed: npm run build.

Notes

• Generated local load-test artifacts were not committed.
• Supabase environments need scripts/schema.supabase.sql applied so new schema fields are available.