feat: refactor schema and path for llama cpp endpoint

[tikikun]

The list of dependency of triton

https://github.com/triton-inference-server/third_party/blob/main/CMakeLists.txt

Drogon can also be a very fast http client and suffer very minimal latency compared to grpc when coupled on the same system.

We should consider not using triton client

[tikikun]

Done alongside with embedding PR