Implement query memoization layer for hot filtered queries

[nicmostert]

Overview

Implement the query memoization layer for whurl. This layer caches results of filtered, expensive queries (e.g., measurement filters) for a short time, to speed up repeated queries and reduce computation.

Acceptance Criteria

• Memoize results of filtered queries derived from canonical caches (sites, measurements).
• Use small, short-lived LRU cache for hot query results (TTL 1-5 minutes).
• Key memoized results by query parameters (function name + normalized filters).
• Do not persist memoized queries across client restarts; memory-only.
• Evict frequently to avoid stale data or cache explosion.
• Expose API for cache clearing/invalidation.
• Document use cases, TTLs, and limitations.
• Comprehensive tests for correctness, cache hits/misses, and performance improvement.