Add initial support for object detection evaluation

[rudrakatkar]

Hi @dpascualhe 👋

This draft PR adds a preliminary object detection pipeline using ImageDetectionModel and ImageDetectionDataset. I downloaded a test image and ran it through a torchvision based maskrcnn model for initial inference and metric testing.
The object detection pipeline works fine and detects the objects in image, but the metrics (precision, recall, and F1) all came back at zero.
I understand this is still tightly coupled with segmentation, so this is just to share how detection evaluation could integrate. I haven’t made any changes to the metrics module yet.
Let me know your suggestions happy to collaborate further on restructuring this properly into the shared architecture discussed in the issue.

Thanks!