Prefix caching

Reusing the KV cache of a prompt prefix shared by requests, so the shared part is not recomputed. The serving-engine counterpart of a provider's prompt caching.

All terms, A to Z