* fix: replace async-openai library with post request to support custom providers * move models call to llm_client too * tweak post processing ui and add groq and cerebras --------- Co-authored-by: CJ Pais <cj@cjpais.com>