Stop Request Stampedes with Gateway Request Deduplication

Platformatic Gateway now includes request deduplication to solve the thundering herd problem. When many concurrent requests hit the same uncached route, only one request goes upstream while the rest wait and receive the same response. The feature uses a leader/waiter model with configurable storage (in-memory or Valkey for horizontal scaling), route whitelisting, custom key functions, and TTL-based deadlock prevention. Benchmarks show 10,000 requests reduced to just 100 upstream calls. Built-in metrics (leader, waiter, replay, fallback counts) let teams verify production impact. Particularly useful for self-hosted Next.js apps handling cache-cold traffic spikes.