[Feat] Socket scale using ip-hash algo

[itssubhodiproy]

This PR introduce the llm-server to scale horizontally by having multiple replicas. Fixes #625

The approach For only socket connections nginx will remember the request ip and redirect all relevant response to the same ip.

Using Nginx

1. REST api requests are getting evenly distributed among available servers.
2. Socket connections stay persistent for ongoing conversation.

Warning: If you wanna test this locally, you will notice that one server (eg. server1) is taking responsibility for multiple chatbots. This is because Nginx is remembering your ip-address and will redirect all socket request to server1. So even-if you're chatting with 100 chatbots in a same time from same ip-address, it will redirect all the requests to that one server.

Solution: It should work in production because the request ip-addresses will be different for users and multiple server instances will come into picture.