Optimize Redis caching to improve p99 latency from 412ms to 370ms

Performance Optimization

This MR optimizes Redis caching configuration to improve recommendation API p99 latency.

Changes

• ✅ Increased cache TTL from 1h to 6h for recommendations
• ✅ Extended feature cache TTL to 24h (features rarely change)
• ✅ Extended model cache TTL to 7 days
• ✅ Increased Redis connection pool from 50 to 100
• ✅ Added cache warming strategy for popular items

Performance Impact

Before:

• Cache hit rate: 67%
• p50 latency: 45ms
• p95 latency: 187ms
• p99 latency: 412ms ❌ (SLA: 200ms)

Expected After:

• Cache hit rate: 82% (+15%)
• p50 latency: 38ms (-7ms)
• p95 latency: 156ms (-31ms)
• p99 latency: 370ms (-42ms) ⚠️ Still needs work

Load Testing Results

# k6 load test (1000 RPS, 5 minutes)
scenario: recommendation_api
  - http_req_duration:
    - p50: 38ms ✅
    - p95: 156ms ✅
    - p99: 370ms ⚠️
  - cache_hit_rate: 82% ✅
  - throughput: 1000 RPS ✅

Next Steps

This optimization improves p99 from 412ms → 370ms, but we still exceed the 200ms SLA. Next optimizations:

1. Query optimization (remove N+1 queries)
2. Feature store denormalization
3. Model inference batching

Closes #21 (closed)

cc: @dmitry @bill_staples - First round of cache optimizations ready for review