feat: add confidence score explanation to output metadata

[Deepak8858]

This PR addresses #416 by:

1. Extending the LandmarkDiffPipeline.generate() output to include confidence and confidence_breakdown fields.
2. Implementing a weighted confidence scoring system based on face detection, identity preservation (ArcFace similarity), landmark accuracy, and mask coverage.
3. Adding an --explain flag to the CLI to print the confidence breakdown.

Fixes #416

[dreamlessx]

Great to see you back @Deepak8858! Fast turnaround on #416.

The _calculate_confidence() approach is clean — weighted average of face detection, identity preservation, landmark accuracy, and mask coverage. The --explain CLI flag is a nice touch.

A few suggestions:

1. landmark_accuracy is currently hardcoded to 0.95. For a follow-up, we could compute this from actual NME when ground-truth landmarks are available.
2. Missing tests — CI will require at least one test for the new method. Something like:

   def test_confidence_breakdown_keys():
       result = pipe.generate(image, procedure="rhinoplasty")
       assert "confidence" in result
       assert set(result["confidence_breakdown"].keys()) == {
           "face_detection", "identity_preservation",
           "landmark_accuracy", "mask_coverage"
       }

3. CI hasn't triggered — this might need a workflow approval in the Actions tab. I'll check.

Not blocking on the hardcoded value — that's a fine v1. Please add a test and we'll merge once CI passes!