fix: merge clip outputs before saving batdetect2 format

[mbsantiago]

Summary

This fixes a regression in batdetect2 process directory when saving predictions in the batdetect2 output format.

When inference produced multiple clip-level predictions for the same recording, each clip was being saved to the same output filename derived from the recording name. That meant later clips overwrote earlier ones, so only detections from the last clip appeared in the final saved file.

This change merges clip-level predictions belonging to the same recording into a single recording-level ClipDetections object before formatting and saving in the batdetect2 format.

Thanks to @Jeff37 for pointing this out!

Additional Note

This branch also includes a small plotting update that was added while diagnosing the issue and is worth keeping.

Testing

• uv run pytest tests/test_inference/test_batch.py tests/test_cli/test_process.py