Renamed some model classes; expanded scope of trivial object classification model to be functional base on intensity threshold, too

Tests parameterize models directly with dicts, not pydantic models, which are reserved for API calls

Put simplest functional RoiSet pipeline model_server for now; derived channel and boundary seg alternatives in trec-adaptive-feedback

Parameters auto-validates anything ending in accessor_id and model_id; call_pipeline() maps these to objects and sends to pipeline function as dicts of respective objects