Enable compatibility with OpenAI-compatible APIs

[thuanha-groove]

This PR enables compatibility with OpenAI-compatible APIs. It also adds the missing nodemon dependency and an ignore rule for locally generated data.

I have tested with:

• LM Studio
• OpenRouter
• Jan

[raidendotai]

• aiming to move all of inference outside of the main api
• to have langsmith-like eval ops that can also trace back to which architecture module (and which architecture) called the inference
• prompts to be at ^ abstraction level , to run benchmarks in batch against prompts

[FeelsDaumenMan]

I tried it with JAN but i dont seem to get it working. A tutorial might be good

[kntjspr]

It's working on OpenRouter but it doesn't work on other OpenAI compatible api like https://glhf.chat/.
It returns 'Failed validation: model must begin with valid prefix (eg. hf:)',

I assume this is also the same case on any huggingface models.