fix mlflow logger

[Zilinghan]

Summary

Update MLFlow logger to better UI display.

Motivation & Context

mlflow downsamples data in UI display (though the full history is never lost) when the x-axis becomes sparse.

Approach

For mlflow, I tried to use per-stage step for different metrics in mlflow logger as the x-axis so that the x-axis becomes less sparse and avoid too much down sampling in the UI.
Also, for very sparse data such as drift.detected, for no detection, there are only logged 10% randomly to avoid to many sparse 0-data-points.

Screenshots / Logs (optional)

API / CLI Changes

N/A

Breaking Changes

• None

Security & Privacy

• No secrets committed
• Input validation added where needed

Dependencies

• N/A

Testing Plan

• Unit tests
• Integration tests
• e2e / smoke test
• Manual steps:
  python -m src.main --config examples/mnist/mnist.toml --set logging.backend=mlflow --set logging.experiment_name="My Experiment"

Documentation

• Docstrings updated
• User docs / README updated
• CHANGELOG entry

Checklist

• Code formatted (Ruff) → ruff format --check
• Lint passes (Ruff) → ruff check .
• Types pass (mypy/pyright) → mypy src
• Tests pass (pytest) → pytest -q

• Backward compatibility considered
• Adequate comments for tricky parts
• CI green

Risk & Rollback Plan

Probably not needed in the beginning

Notes for Reviewers

FYI, merging this to the main branch first may cause some conflict when we merge the #83 PR later