Problem

The semantic search functions (match_content_embeddings) were taking much longer than the underlying queries. This performance degradation was caused by the LIMIT being passed as a parameter to the function body, which prevented the PostgreSQL query planner from optimizing the query effectively.

Solution

This PR implements the optimization suggested in ENG-1769:

1. Removed match_count parameter from match_content_embeddings function
2. Removed LIMIT clause from the function body to allow the query planner to optimize properly
3. Moved LIMIT to callers using Supabase's .limit() method on the client side

Changes

Database Changes

• Updated match_content_embeddings function signature:
  • Removed: match_count integer parameter
  • Removed: LIMIT match_count clause from function body
• Created migration file to document the change

Application Changes

• apps/roam/src/utils/hyde.ts: Updated findSimilarNodesVectorOnly to use .limit() method
• apps/roam/src/utils/discourseNodeSearchProviders.ts: Updated semantic search to use .limit() method
• packages/database/src/dbTypes.ts: Updated TypeScript types to reflect new function signature

Performance Impact

By removing the LIMIT from the function body, the PostgreSQL query planner can now:

• Better optimize the vector similarity search
• Make more informed decisions about index usage
• Potentially use parallel query execution

The function should now have roughly the same performance as running the underlying query directly.

Testing

The changes maintain backward compatibility in terms of behavior:

• All existing callers have been updated to apply the limit on the client side
• The results returned will be identical to before, just with better performance

Linear Issue: ENG-1769