Filter invalid bounding boxes in instance segmentation

[guarin]

What has changed and why?

• Apply filtering as in object detection

How has it been tested?

• Manually

Did you update CHANGELOG.md?

• Yes
• Not needed (internal change)

Did you update the documentation?

• Yes
• Not needed (internal change without effects for user)

[guarin]

/review

[Copilot]

Pull request overview

This PR adds filtering of invalid bounding boxes in instance segmentation transforms, applying the same filtering pattern already used in object detection models. This ensures that bounding boxes that become invalid after transformations (e.g., due to cropping or rotation) are properly filtered out.

Changes:

• Added conditional bbox filtering parameters (filter_invalid_bboxes and clip) to instance segmentation train transforms, matching the pattern used in object detection
• Updated CHANGELOG.md to document the fix

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/lightly_train/_task_models/dinov3_eomt_instance_segmentation/transforms.py	Added version-conditional bbox filtering parameters (filter_invalid_bboxes for albumentations>=2.0.1 and clip for albumentations>=1.4.5) to DINOv3EoMTInstanceSegmentationTrainTransformArgs
CHANGELOG.md	Added entry documenting the fix for invalid bounding box filtering in instance segmentation

[liopeer]

LGTM!