Task 3b: Add solar-panel-defects detection use case

[johyeongseob]

Summary

This PR adds the Task 3b Dataset A1 solar_panel_defects use case based on the PVEL-AD electroluminescence image dataset.

Changes

• Add the solar_panel_defects use-case configuration, policy, prompts, and predefined questions.
• Add OpenVINODetectHandler for single-channel YOLO-style OpenVINO detection output.
• Decode [1, 16, 8400] output and apply confidence filtering and class-aware NMS.
• Restore bounding boxes to the original image coordinates.
• Store detection results in JSONL and SQLite with modality_source: electroluminescence.
• Integrate the standard policy, analysis, and evidence agents.
• Add CLI and Web UI chat support.
• Add handler tests and use-case documentation.

Validation

• 13 passed across Solar Panel Defects and Oil & Gas regression tests.
• Verified inference on 100 PVEL-AD test images:
  • 217 raw detections
  • 131 policy-kept detections
  • 86 filtered detections
• Verified Analysis, Evidence, and SQL chat modes in both CLI and Web UI.
• Evaluated the full 19,150-image PVEL-AD test set:
  • Precision: 0.8789
  • Recall: 0.7822
  • F1: 0.8277

Notes

The PVEL-AD dataset, OpenVINO model files, and local evaluation script are excluded from this PR.

Thank you,
Hyeongseob Jo

[rv355]

Avoid hard-coded components in the agent.

[johyeongseob]

Thanks for the feedback.

I removed the hard-coded use-case logic and moved the report rules to each use-case prompt. All tests passed.