[FEAT]: Model router

[shatfield4]

What would you like to see?

• Build an intelligent prompt routing layer that intercepts user prompts before they reach LLM to dynamically select the model that would provide the best result for the costs
• Main idea is to use calculated (deterministic) rules such as: keyword triggers, token count, message counts, date/time periods (some models have peak pricing times) etc
• See if we can get semantic routing working by allowing users to configure a specific model used for routing chats inside admin settings for model router and using that model to do a tool call to route to the correct model each time
• Follow backend design per internal spec on Notion

[PikkonMG]

In the works I asked a few weeks ago #5209