Thread-Safe Inference Result Cache

Arrowstreet Capital's model inference service repeatedly requests results for the same input features. Implement a fixed-capacity, thread-safe least recently used (LRU) cache that stores pre-computed inference results.

The cache must support get and put operations. A successful get marks the key as most recently used. When inserting a new key would exceed capacity, evict the least recently used key. Updating an existing key also makes it most recently used. Every cache operation must be protected so concurrent callers cannot corrupt the cache's internal state.

Formal Specification

Implement cache_inference_results(capacity, operations).

• capacity is an integer greater than zero.
• operations is a list of dictionaries. Each dictionary is either {"op": "get", "key": string} or {"op": "put", "key": string, "value": integer}.
• Process operations in the supplied order and return a list containing the result of every get operation.
• Return the stored integer for a hit and None for a miss.
• The implementation must use a lock around shared cache state. The returned order follows the input operation order, while the cache class itself must be safe for concurrent callers.