feat: switch to custom FastAPI app to avoid server issues

fix: remove unsupported log-level flag

chore: add debug logging to diagnose 400 errors

feat: set model alias for OpenAI API

fix: use original HF model for tokenizer

deps: add sentencepiece for tokenizer

deps: add transformers for HFTokenizer

fix: remove --api flag (misinterpreted as --api_key)

chore: use prebuilt wheels to avoid compile time

Fix: enable OpenAI API with --api flag and tokenizer path

Update Dockerfile

Create Dockerfile