Cease support for llama.cpp-served Gemma

[AstraBert]

Reference to #2 but also to the inefficiency of the solution

Explore new local serving methods like quantization (non-dockerizabble) and llama.cpp python package