Cache: add new flag to distinguish models that Cache but not static cache

[gante]

What does this PR do?

See title :)

Models like Jamba (will only ever support HybridMambaAttentionDynamicCache) or Mistral (currently only supports dynamic caches, like SinkCache) need a distinction in terms of support flags, to run the appropriate checks and tests.

Without this flag, we have the following bad practices:

1. manually skip tests
2. don't do cache support checks in generate

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[zucchini-nlp]

Thanks for fixing the tests!

[ArthurZucker]

Alright, not super fan of increasing the things a model supports by adding more attributes, but fine for now.

[gante]

@ArthurZucker we can remove this attribute when we update all Llama-copied models to support the static cache :D