ObjectDetectionTask: increase test coverage for torchvision 0.14+

[isaaccorley]

The try/except in the ObjectDetection trainer's validation step was suppressing the following errors which drives this PR. These errors are caused by the use of torch.rand to generate dummy bounding box data which results in some cases where xmax < xmin and/or ymax < ymin which results in the bounding box plotting throwing a ValueError

FAILED tests/trainers/test_detection.py::TestObjectDetectionTask::test_trainer[True-faster-rcnn-nasa_marine_debris] - ValueError: Boxes need to be in (xmin, ymin, xmax, ymax) format. Use torchvision.ops.box_convert to convert them
FAILED tests/trainers/test_detection.py::TestObjectDetectionTask::test_trainer[True-fcos-nasa_marine_debris] - ValueError: Boxes need to be in (xmin, ymin, xmax, ymax) format. Use torchvision.ops.box_convert to convert them
FAILED tests/trainers/test_detection.py::TestObjectDetectionTask::test_trainer[True-retinanet-nasa_marine_debris] - ValueError: Boxes need to be in (xmin, ymin, xmax, ymax) format. Use torchvision.ops.box_convert to convert them

Additionally, this PR adds a missing test for covering the case where fig returned by the plot method is None because the dataset has no plot method.

[adamjstewart]

Curious why we never encountered this issue before, but it fails consistently every time in your PR that bumps torch/torchvision versions. Did something change?

There's a lot of code duplication here. Can we:

1. Move plot_no_rgb and no_plot_method to tests/trainers/conftest.py so they can be shared?
2. Combine test_no_rgb and test_no_plot_method into a single parameterized test_plot method?

[isaaccorley]

@adamjstewart, I may not have time to finish this in the next few days since I'm traveling. Not sure if you want to finish it up.

[adamjstewart]

Did something change?

Something did indeed change: pytorch/vision#6123

So we used to be generating (10, 4) random floats. The chances that any particular tuple satisfies xmax > xmin AND ymax > ymin is 0.25. The chances that all 10 tuples satisfy this is $0.25^{10} = 9.5^{-7}$. This explains why the tests basically always skipped that line in torchvision 0.14+.

[adamjstewart]

Thanks for finding this bug! Glad it was only in our testing code and not in the core library.