Added task definition and configs for Panoptic Mask R-CNN (#10165)

* added PanopticMaskRCNNTask config

* added tests for config

* added Panoptic MaskRCNN task

* added tests for Panoptic MaskRCNN task

* fixed typo

* set default segmentation loss weight to 1.0

* added class docstrings

* support `segmentation_decoder` checkpoint

* added `init_checkpoint_modules` description

* compute `steps_per_epoch` with `train_batch_size`

* flattened semantic_segmentation loss attributes

official/vision/beta/projects/panoptic_maskrcnn/configs/panoptic_maskrcnn.py

@@ -47,13 +47,11 @@ class Parser(maskrcnn.Parser):
       default_factory=list)
   segmentation_ignore_label: int = 255
 
-
 @dataclasses.dataclass
 class DataConfig(maskrcnn.DataConfig):
   """Input config for training."""
   parser: Parser = Parser()
 
-
 @dataclasses.dataclass
 class PanopticMaskRCNN(maskrcnn.MaskRCNN):
   """Panoptic Mask R-CNN model config."""