Consider using Bifrost as OpenAI-compatible LLM gateway server

Consider using bifrost's core https://github.com/maximhq/bifrost instead of re-implementing the gateway server logic.

P.S. I'm not affiliated with Bifrost, my motivation is because I'm already using Bifrost and consider using Jan Server too, so making them natively work in tandem would be nice.