updated to faster models, now uses mini, mini, and haiku

storing the audio_bytes in the interviewState rather than passing it back to a file structure.

setting up docker to be able to deploy the in streamlit app.

first node is working!