Switch back to subprocessing for llama.cpp

This should resolve a number of memory leak and stability defects by allowing
us to isolate llama.cpp in a separate process and shutdown when idle, and
gracefully restart if it has problems.  This also serves as a first step to be
able to run multiple copies to support multiple models concurrently.

llm/generate/generate_darwin.go

@@ -1,3 +1,3 @@
 package generate
 
-//go:generate sh ./gen_darwin.sh
+//go:generate bash ./gen_darwin.sh