Implement wrappers around any augmentations in imgaug or albumentations

The functionality of our augmentation interface should be strong enough to support a generic wrapper that can wrap any augmentations in imgaug/albumentations into a subclass of our Augmentation.

Some reference:

• Wrappers from tensorpack: https://github.com/tensorpack/tensorpack/blob/8e6f930b5f021c21f785619278e7ec9beacc0a1b/tensorpack/dataflow/imgaug/external.py#L11-L63
• doc of detectron2 augmentation: https://detectron2.readthedocs.io/en/latest/tutorials/augmentation.html

Interface should look like:

from detectron2.data import transforms as T
import albumentations as A
T.Albumentations(
  A.GaussianBlur(),
  A.xxx
)

[KUASWoodyLIN]

#3609 Hi I have implement albumentations wappers, it have image, bounding box and segmentation augmentation supported.

[Kevin-Delnoije]

@SnowRipple you can easily copy paste datasetmapper.py and change L 134-136 with any other augmentation. Bit of a hacky solution so i hope this issue is still being worked on!