feat: implement read replicas, queue-based math updates, and fix scal…

[nicobao]

…ability issues

Read Replica Infrastructure:

• Add read replica support with automatic query routing via withReplicas()
• Primary DB for writes, replica for SELECTs across API and math-updater
• AWS RDS configuration with Secrets Manager support

Queue-Based Architecture:

• Replace needsMathUpdate flag with conversation_update_queue table
• Rate limiting based on lastMathUpdateAt (actual processing time)
• Race-condition safe updates with requestedAt timestamp matching

Counter Reconciliation:

• Self-healing counters reconciled by math-updater every 20s
• Eliminates expensive COUNT queries from opinion/vote hot path
• Updates lastReactedAt to track conversation activity

Math-Updater Scalability Fixes:

• Upgrade pg-boss v10→v11 with 'singleton' queue policy
• Prevent concurrent job execution for same conversation
• Fix Drizzle ORM stack overflow with per-batch CASE statements (1000/batch)
• Handle 113k+ votes, 19k+ opinions without crashes (50-85s processing time for backend math update, frontend runs smoothly all the way)

Load Testing:

• Add k6 load testing with 2 scenarios (single/multiple conversations)
• Grafana dashboards + Prometheus monitoring for PostgreSQL
• UCAN crypto implementation for authenticated k6 tests

Performance Impact:

• Eliminated database lock contention causing system hangs under load
• Sequential math processing (no concurrent waste)
• Self-healing counters accurate within 20 seconds
• System responsive under 200 (very active) concurrent users per conversation

Documentation:

• Comprehensive performance analysis with load test results
• Counter reconciliation strategy documentation
• AWS RDS deployment guide with read replica architecture

Python-bridge

• rollback to previous version of red-dwarf that didn't have a bug that prevented math update
• improve logging