Add typing annotations to models/detection

🚀 The feature

This is an issue to track down all the files in detection module and work up the chain.

• _utils.py #4583
• anchor_utils.py #4599
• backbone_utils.py #4603
• image_list.py #4602
• roi_heads.py #4612
• rpn.py #4619
• generalized_rcnn.py #4631
• transform.py #4630

To type the models, all the above files would need to have annotations.

• faster_rcnn.py #4636
• mask_rcnn.py
• keypoint_rcnn.py
• retinanet.py
• ssd.py
• ssdlite.py

Motivation, pitch

@khushi-411 and I will sync and work on these files.

Solving This issue also helps #2025

As first step I'm closing #4220 and splitting it up into multiple PRs preferably one file each.

Hopefully we should complete this soon.

Alternatives

Note that there are cases where we have make JIT happy over mypy.

Most of these cases occur as JIT does not Support Tuple fully. In most cases we should prefer List.

Additional context

cc @datumbox @pmeier @NicolasHug

Review the unchecked files listed in the issue, including anchor_utils.py, roi_heads.py, transform.py, faster_rcnn.py, mask_rcnn.py, keypoint_rcnn.py, retinanet.py, ssd.py, and ssdlite.py, alongside the completed-file checklist. Start by comparing the existing annotation work in the checked files and account for the stated JIT-versus-mypy constraint. Done means the detection model files are annotated and the related work is complete.