[MLOps] Wrap model scripts with MLflow experiment tracking

[ayushcmd]

Currently all 4 model scripts (XGBoost, LSTM, Transformer, meta-ensemble)
run independently with no experiment tracking.

What needs to be done

• Wrap each script with MLflow start_run()
• Log parameters, metrics, and artifacts for every run
• Store runs under named experiments (e.g. xgboost-dsi, lstm-forecast)
• Add mlops/ folder to keep tracking code separate from src/

Why

Makes it easy to compare model versions, reproduce results,
and eventually plug into a Model Registry.

Part of

MLOps layer planned for Dollar Hegemony v2