power with llama cpp under the hood instead of ollama

[nischalj10]

see for more details - nischalj10/headless-ollama#1

[nischalj10]

update - llama.cpp has stopped supporting VLMs on its server. ggml-org/llama.cpp#5882