feat(cache): add tiered query cache with in-memory and optional Redis backends

[ponderingdemocritus]

Summary

Adds a query result caching layer to improve read performance for frequently accessed data.

What this PR adds:

• Query cache module (torii-cache): New caching infrastructure with a QueryCache trait and multiple backend implementations
• In-memory cache: Thread-safe cache using DashMap with TTL-based expiration (always available)
• Redis cache: Optional distributed cache backend for multi-instance deployments
• Tiered caching: Tries Redis first, falls back to in-memory - writes to both for redundancy
• CachingPool: Wrapper around SQLite pool that intercepts SELECT queries and caches results
• Cache key generation: SHA-256 hash of SQL + bind parameters, with table name extraction for pattern-based invalidation
• Automatic invalidation: Caches are invalidated by table pattern when data is modified

CLI options:

--cache.enabled      Enable query result caching (default: false)
--cache.ttl          Cache TTL in seconds (default: 60)
--cache.redis_url    Optional Redis URL for distributed caching

Integration:

• gRPC SqlQuery endpoint now uses cached queries via CachingPool
• Write operations trigger pattern-based cache invalidation (e.g., torii:query:entities:*)

Test plan

• cargo check -p torii-sqlite -p torii-grpc-server -p torii-runner
• Unit tests for cache backends (cargo test -p torii-cache)
• gRPC integration tests (currently blocked by katana timeout / db version mismatch)