Add support for caching using PSR-6 #141
Description
There are moments where Providers will want to cache results and not make a request every time. An example of this is that our current Provider implementations request the list of models for every fulfilled prompt. This means every prompt is at least two round trips when it could be one in subsequent calls.
We should implement the PSR-6 interfaces to make this work. This allows any system implementing this to provide its own caching layer.
I did consider the PSR-16 route, especially since maps pretty well to the wp_cache_* functions in WordPress, for us in the WP AI Client, but in the spirit of keeping this more framework agnostic, I think we should use PSR-6. It's really not much more complex, but allows for more verbose caching systems.