prefix caching in llama.cpp

[bytesupply76]

Spent some time on this today and wanted to compare notes.

Two paths both seem viable, but they break down differently once VRAM headroom becomes important.

Trying to avoid rebuilding the whole thing twice.