Implement Circuit Breaker and Bulkhead Pattern for Redis Cache Backend

[SanshayMaestro]

PR Description — Implement Circuit Breaker and Bulkhead Pattern for Redis Cache Backend

Summary

This PR hardens the Redis cache backend against service degradation and connection failures by introducing a circuit breaker, exponential backoff health checks, in-memory fallback, and bulkhead isolation.

The goal is to prevent Redis failures from cascading into application-wide failures, reduce thundering-herd behavior during recovery, and prevent Redis connection exhaustion.

Changes

• Implemented a configurable Closed → Open → HalfOpen circuit breaker.
• Added exponential backoff for Redis health checks with a 100ms base delay and 30s maximum delay.
• Added fallback to InMemoryCache when Redis is unhealthy or the circuit is open, subject to the configured TTL.
• Added bulkhead controls to limit concurrent Redis operations and queue depth.
• Added circuit-breaker error variants and configuration options.
• Added metrics for circuit transitions, rejected operations, and recovery attempts.
• Added integration coverage for graceful degradation during Redis failures.

Affected Areas

• src/cache.rs — Redis health checks, connection handling, fallback, and isolation.
• src/config.rs — Circuit breaker configuration.
• src/error.rs — Circuit breaker-related error variants.

Testing

Integration tests verify that the cache backend:

• Opens the circuit after the configured failure threshold.
• Transitions to HalfOpen and recovers when Redis becomes healthy.
• Uses exponential backoff during health checks.
• Falls back to in-memory caching during Redis outages.
• Rejects operations when bulkhead limits are exceeded.
• Recovers gracefully after Redis availability is restored.

Expected Outcome

Redis degradation should no longer cause cascading failures across the application. The cache layer should remain available through controlled fallback behavior while preventing excessive retries, thundering-herd effects, and connection exhaustion.
Closes #54

[MaryammAli]

LGTM
a job done and well implemented