torchvision_finetuning_instance_segmentation.ipynb updates

Issue 1:

On following the colab notebook posted here, the run fails

ValueError                                Traceback (most recent call last)
[<ipython-input-19-b563ca732476>](https://83yy92n3d-496ff2e9c6d22116-0-colab.googleusercontent.com/outputframe.html?vrz=colab-20230302-060133-RC02_513678701#) in <module>
      4 # let's adda color palette to the mask.
----> 5 mask.putpalette([
      6     0, 0, 0, # black background
      7     255, 0, 0, # index 1 is red

2 frames
[/usr/local/lib/python3.8/dist-packages/PIL/Image.py](https://83yy92n3d-496ff2e9c6d22116-0-colab.googleusercontent.com/outputframe.html?vrz=colab-20230302-060133-RC02_513678701#) in _getdecoder(mode, decoder_name, args, extra)
    431     except AttributeError as e:
    432         raise OSError(f"decoder {decoder_name} not available") from e
--> 433     return decoder(mode, *args + extra)
    434 
    435 

ValueError: unknown raw mode for given image mode

Issue 2:

Path to MaskRCNN and FasterRCNN model architectures do not load

Fix:

1. Enable P mode in mask before calling putpalette()
2. Update image paths in colab

Since I do not have access to the original notebook and am unable to find the source, I have made a copy and updating the link in tutorial.

Start with intermediate_source/torchvision_tutorial.rst and follow its linked Colab notebook. Check the mask setup around putpalette() and the MaskRCNN and FasterRCNN image paths; done means the notebook loads the models and runs past the palette error with the updated link and paths.