Add matplotlib image support

[brollb]

Currently, deepforge supports using matplotlib for generating 2D plots. It would be great if this could be extended to support displaying images as well (particularly useful for image classification problems!).

[brollb]

Job feedback is currently encoded in the stdout for a job, decoded by the server, and then the project is updated to contain the actual corresponding objects. As matplotlib supports custom backends, we have a custom backend defined for use within deepforge which prints a JSON representation of the graph when using it for plotting.

To add support for images, we would need to:

• save the image to the filesystem (add to the custom matplotlib backend)
• upload the image to the deepforge server (using the blob client and retrieving the hash)
  • This should already be done here as long as the matplotlib backend saves the image to metadata/ as a png and then prints the cmd to stdout.
  • An example can be found in deepforge v1 where this was performed by a deepforge global in lua: https://github.com/deepforge-dev/deepforge/blob/v1.x/src/plugins/GenerateJob/templates/deepforge.ejs#L80-L81
• encode an image update message in the stdout which passes the image info and the hash
  • This should already be handled by the link from the previous point
• ensure that the image update message is decoded properly by deepforge and the corresponding image node is created.
  • This should work out of the box but there may have been changes since v1 that may require some modifications (namely here)

[umesh-timalsina]

Now that #1329 is merged, Incremental support to Image trace with updated meta-model can be provided.

[umesh-timalsina]

Steps for Image Support is quite different for plotly than #1218 (comment).

1. Custom Matplotlib Backend Should be updated to get Image Information(The best strategy now is the JSON Dump a numpy Matrix although this could be changed to support any other form of encoding).
2. Update The MetaModel such that node Image is contained within the SubGraph.
3. So, the image information will now be available upon plots creation.
4. Update the ExecuteJob.Metadata.js to create Image Nodes upon getting the information.
5. Update PlotlyDescExtractor.js to get the plotly JSON from the image Node.
6. Update ExecutionIndexControl.js to send updates regarding Images to the Widget.

1. Custom Matplotlib Backend Should be updated to get Image Information(The best strategy now is the JSON Dump a numpy Matrix although this could be changed to support any other form of encoding).

This strategy works but since the dumped matrix is very large, stdout for a execution is almost full of the Matrix. Rather than dumping this array, this should be converted to png and uploaded to the file system borrowing some ideas from #1218 (comment)

Another Alternative to this strategy is to Base64Encode the matplotlib Image numpy Array.