idea: Support more parameters for /v1/models/start

[vansangpfiev]

Problem Statement

Currently, /v1/models/start only accept prompt_template parameter

Feature Idea

Should accept more parameters like context_length
Scenario:

• When the user sends start request, cortex will override parameters loaded from model.yml by request's parameters.
• The content of model.yml will be kept as-is.

[vansangpfiev]

Need to add below parameters:

cache_enabled: bool
ngl: int
n_parallel: int
cache_type: string
ctx_len: int

[vansangpfiev]

To support vision model, we also need to add mmproj parameter, type: string

mmproj: string
llama_model_path: string
model_path: string

[gabrielle-ong]

API: accept parameters (though it accepts any parameter - is there some guardrail)

Eg test n_parallel