Compatibility of demo code with the new baselines for Instance Segmentation with Mask R-CNN

## 🚀 Feature
Adding the support for the new baseline instance segmentation models introduced in [Model Zoo ](https://github.com/facebookresearch/detectron2/blob/master/MODEL_ZOO.md) for the `demo.py`. Currently the code is written so that it only supports checkpoints with `.yaml` configs and models with Yacs Config . I have been able to load the checkpoints using [Lazyconfigs](https://detectron2.readthedocs.io/en/latest/tutorials/lazyconfigs.html) instructions, however, I have not been able to populate the detector model with the loaded weights.

I am using detectron2 as an off the shelf object detector in the system I am designing. Therefore, I am not doing any training for the object detector and it would be very nice if I can easily use your demo code with the new baseline models.

Start in demo.py and compare its existing YAML/Yacs checkpoint path with the LazyConfig loading instructions linked in the issue. Trace how the loaded checkpoint is expected to populate the detector model, then verify that the new Model Zoo instance-segmentation baselines can run through the demo code with their weights loaded.

Written by the indexing model from the issue text.