feat: cortex support custom prompt_template when load model

[nguyenhoangthuan99]

This feature supports Jan or other applications that want to integrate Cortex as a server. When users want to change the prompt template, Cortex can directly load that prompt instead of saving it to model.yml and then loading the model from model.yml.

[gabrielle-ong]

QA: ✅

• API shouldn't save the prompt template to model.yaml file (this case is handled by POST models/update)
• try inputing a random prompt template, model should still load successfully (optional parameter)
• usecase is requested by Louis when he intergrating Jan
• to NOT save the prompt template when loading model if user accidentally modifies prompt template during models/start