poor scaling of `functools.lru_cache`

functools.lru_cache currently scales poorly in free-threading because of reference counting contention on the object. The underlying function has deferred reference counting enabled but the cache object itself needs deferred reference counting as well to perform well in free-threading.

Linked PRs

• gh-142632

首先检查 functools.lru_cache 的实现和相关 PR gh-142632，然后研究 free-threading 如何影响 cache 对象的引用计数。当 cache 在 free-threading 下具有更好的扩展性，同时保留 lru_cache 现有的行为和测试时，即可视为完成。