add disclaimer and ui updates

fix: sanitize PIL images to numpy arrays, add python_version to configuration block

implement environment-driven model loading from Hugging Face

add streamlit app

implement unified batch pipeline and clean up obsolete scripts

build YOLOv8 data pipeline and configure diatom detector

train V2 model with augmented data, class weights, and update inference default

add 15% margin to bounding box crops to prevent clipping

add CLI inference script with lazy loading and usage documentation

initial project setup and data extraction script