[feat] Add pluggable CircuitBreakerStorage interface with Redis adapter for multi-instance deployments #50

Description:

Problem

In serverless environments (AWS Lambda, Vercel Functions) and multi-pod Kubernetes clusters, the circuit breaker state is stored in process memory (CircuitBreakerState).
If 50 pods are running and an upstream service goes down, each pod must independently experience failureThreshold errors before opening its circuit. This results in hundreds of doomed requests hammering an already failing upstream service.

Proposed Solution

1. Abstract state management behind a CircuitBreakerStorage interface:
   • getState(domain: string): Promise<CircuitEntry>
   • recordFailure(domain: string, threshold: number, cooldownMs: number): Promise<CircuitState>
   • recordSuccess(domain: string): Promise<CircuitState>
2. Provide a default InMemoryStorage (current behavior).
3. Provide an optional RedisCircuitBreakerStorage (or Redis Lua script) for atomic sliding window failure counters across distributed instances.
4. Add resilience so that if Redis becomes unreachable, it falls back to local memory without failing requests.

Acceptance Criteria

• Storage interface defined and decoupled from CircuitBreakerState.
• Redis storage implementation provided with atomic increments/state checks.
• Graceful fallback to in-memory if Redis connection fails.
• Integration tests verifying state synchronization across separate client instances.