Extending panoptic segmentation support to more image formats

## 🚀 Feature
Support more image formats other than JPG (e.g. PNG and BMP)

## Motivation & Examples

For example, when loading panoptic annotations the user is forced to have images in JPG format, the only way to change the behaviour is to change the code. Moreover having images in JPG format is detrimental since it adds artifacts to the images that should be used to train models.

https://github.com/facebookresearch/detectron2/blob/f755c495a407ea8633d0eb871abc4693c2d7b1b4/detectron2/data/datasets/coco_panoptic.py#L49