Enable to download checkpoint from url for inference

[georgeyiasemis]

Main changes:

• DefaultConfig allows for checkpoint_url input. Should be a url that downloads model weights that match with the model configuration in the DefaultConfig
• If url does not correspont to a torch state_dict this will throw an error and exit.
• If --checkpoint <checkpoint_num> is passed in the predict_val.py and predict_test.py whilist checkpoint_url is not null, --checkpoint <checkpoint_num> will be ignored