feat(cache): add cache store abstraction layer

Implement reusable cache store abstraction with in-memory and Redis
backends as foundation for prompt caching feature (PR1 of progressive delivery).

- Add CacheStore protocol defining cache interface
- Implement MemoryCacheStore with LRU, LFU, and TTL-only eviction policies
- Implement RedisCacheStore with connection pooling and retry logic
- Add CircuitBreaker for cache backend failure protection
- Include comprehensive unit tests (55 tests, >80% coverage)
- Add dependencies: cachetools>=5.5.0, redis>=5.2.0

This abstraction enables flexible caching implementations for the prompt
caching middleware without coupling to specific storage backends.

Signed-by: William Caban <willliam.caban@gmail.com>

pyproject.toml

@@ -26,6 +26,7 @@ classifiers = [
 dependencies = [
     "PyYAML>=6.0",
     "aiohttp",
+    "cachetools>=5.5.0",                              # for prompt caching
     "fastapi>=0.115.0,<1.0",                          # server
     "fire",                                           # for MCP in LLS client
     "httpx",
@@ -37,6 +38,7 @@ dependencies = [
     "python-dotenv",
     "pyjwt[crypto]>=2.10.0",                          # Pull crypto to support RS256 for jwt. Requires 2.10.0+ for ssl_context support.
     "pydantic>=2.11.9",
+    "redis>=5.2.0",                                   # for prompt caching (Redis backend)
     "rich",
     "starlette",
     "termcolor",