Remote evaluation doesn't work

[rcannood]

Running eval.ai with docker-compose and trying to connect with a remote worker does not work. It seems the remote worker script is a bit out of date with the current API. The API documentation helped out a lot in figuring out what the issue was.

These are the issues I had to fix to get the script to work:

• Rename Token to Bearer (L329).
• Remove return statement (L527). I don't understand why this is there.
• Fix the submission_data["result"] data structure format. The evaluation script is expected to return an object in the format of :

  [ 
    { "train-split" : { 
      "Metric" : 100 
    } } 
  ]
  

  but the API expects :

  [ 
    {
      "split": "train-split", 
      "show_to_participant" : True, 
      "accuracies" : { "Metric" : 100 } 
    }
  ]
  

  So a little bit of post-processing is needed.

I will update #3496 to contain these fixes.

@Ram81: After I update the PR, can you review my changes?