Docs: add MODEL_CARD.md documenting ML weights provenance

[Alimedhat000]

Summary

The ML model weights' provenance is undocumented. The model is downloaded at worker startup from Google Drive via gdown using GDRIVE_WEIGHTS_ID, but there is no record of what dataset it was trained on, how it was validated, or who built it.

Evidence

• docker/worker-entrypoint.sh — downloads weights/*.pth via gdown "$GDRIVE_WEIGHTS_ID" --folder if absent
• GDRIVE_WEIGHTS_ID hardcoded in docker/docker-compose.yml, docker/docker-compose.dev.yml
• No MODEL_CARD.md or equivalent anywhere in the repo
• ml/_FINAL_SCRIPTS/ contains the inference scripts and weights, but no training/validation documentation

Suggested fix

Add a MODEL_CARD.md (or a docs note) covering, per model (tremor, drawing/spiral, voice, questionnaire):

• source dataset(s) and their licensing
• training procedure (base architecture, augmentation, split)
• validation approach and reported metrics
• model I/O contract (input file formats, expected score semantics)

Note

This is also interview-relevant: "where did the model come from / how was it validated" is a likely question for the resume project this backs.