Automatic cache clearing with lru_cache's maxsize

[Pierre-Sassoulas]

Following #1139, it's not possible to clear the cache. We did not use lru_cache because the current implementation only uses the first argument as key (because the other one is context?), and lru_cache does not allow that.

In order to use lru_cache we need to ask ourselves if is it right that inference tip functions always take exactly two arguments -- a node and a context? If so, what is context exactly? Does it remain constant during an inference session? Is it fine to hash it via id()?

See original comment by @superbobry here and here

Current behavior

The cache can be cleared with an API.

Expected behavior

Automatic clearing when the cache increase above a set value (configurable ?).

[jacobtylerwalls]

Reopening because #1565 will likely be reverted.

[jacobtylerwalls]

The interface for clearing astroid's various caches is astroid.MANAGER.clear_cache(). If we missed one, please open a new issue and provide details. Thanks all!