Architect stale cache probabilistic eviction for high-throughput indexer API

[El-swaggerito]

Closes #601

I've successfully implemented a stale cache probabilistic eviction mechanism for Web3 webhook ingestion. Here's what was delivered:

Implementation Summary

1. Stale Cache Service (src/services/staleCache.ts)

• Implements stale-while-revalidate pattern with probabilistic eviction
• Returns stale data immediately to avoid blocking the event loop
• Probabilistically refreshes cache in background based on access frequency
• Integrates with existing Count-Min Sketch for frequency tracking
• Configurable via environment variables

2. Webhook Service Integration (src/services/webhookService.ts)

• Integrated stale cache for webhook configuration lookups
• Non-blocking reads during high load scenarios
• Automatic cache invalidation on configuration updates
• Maintains backward compatibility with existing Prisma schemas

3. tRPC Middleware (src/trpc/staleCacheMiddleware.ts)

• Type-safe middleware for stale cache operations
• withStaleCache() for query procedures
• withCacheInvalidation() for mutation procedures
• Maintains absolute type safety across tRPC boundary

4. Comprehensive Testing (src/services/tests/staleCache.test.ts)

• Event loop non-blocking tests
• High throughput tests (100+ concurrent operations)
• Probabilistic refresh behavior validation
• Error handling under load
• Memory leak prevention tests
• Input validation edge cases

Acceptance Criteria Met

✅ Non-blocking asynchronous architecture: All cache operations return immediately, with background refreshes that don't block the event loop

✅ Type safety across tRPC boundary: Middleware maintains full TypeScript type safety with generic type parameters

✅ Backward compatibility: Integration with existing Prisma schemas without modifications

✅ Edge case handling: Comprehensive input validation, error recovery, and graceful degradation

✅ Performance under load: Tests demonstrate handling 100+ concurrent operations without event loop blocking

Configuration

Add to your .env file:

STALE_CACHE_THRESHOLD_MS=30000
STALE_CACHE_EXPIRE_MS=300000
STALE_CACHE_REFRESH_PROB=0.3
STALE_CACHE_DEFAULT_TTL=60
STALE_CACHE_ENABLE_REFRESH=true

The implementation leverages your existing probabilistic eviction infrastructure (Count-Min Sketch) and Redis caching layer, ensuring minimal overhead while providing significant performance improvements during heavy Web3 block finalization spikes.