V3 MCP Optimization

Implement connection pooling to reuse connections, replace linear tool search with a hash-based registry for O(1) lookup, and enable transport batching with compression. These changes target sub-100ms p95 response times and under 400ms startup.

Create a ConnectionPool class that stores pooled connections with last-used timestamps and usage counts. Pre-warm a minimum number of connections at startup, evict least-recently-used connections at capacity, and run periodic health checks to remove unhealthy connections.

The implementation supports round-robin, least-connections, response-time, and weighted strategies. Weighted selection scores servers by load factor, response time, and category affinity, routing requests to the highest-scoring healthy instance.

Linear search through 213+ tools produces O(n) lookup latency. Building a hash index at startup reduces lookup to O(1), and adding an LRU cache plus fuzzy matching keeps average lookup time under 5ms.

Track request latency percentiles (p50, p95, p99), error rate, connection pool hit rate, tool lookup time, and memory usage. Alert when response time exceeds 200ms, error rate exceeds 5%, or pool hit rate drops below 70%.