separate out model_server and remove dependencies of model server from archgw pip package

[adilhafeez]

Currently model_server cli is built and installed as part of archgw pip package. This results in pulling in many dependencies that are not directly related if we are just using llm_routing (egress_routing). Splitting model server off from archgw will make archgw cli a bit simpler by not having to serve models directly from archgw cli. This would require some thinking and planning on how to effectively work on to separate out model_server.

Ideally archgw cli will manage llm_gateway and prompt_gateway and whenever it needs access to local or remote hosted models we'd need hosted model endpoint.

Following models would need to be hosted and managed outside of archgw cli,

• arch-routing model (llm_routing)
• arch-function model (prompt_gateway)
• guard-model

[junr03]

This is more involved and not directly involved in my expertise, but I can take a stab at it after #513 #514 #515. Does that sound fair?