GitLab

If someone checks out the ollama repo and doesn't install the CUDA
library, this will ensure they can build a CPU only version

Run the server.cpp directly inside the Go runtime via cgo
while retaining the LLM Go abstractions.