Add instructions on how to submit a model

[gwenzek]

Note that this PR has external dependencies:

• releasing the latest changes to Dynabench for GPU support ( @douwekiela )
• merging changes to Dynalab https://github.com/facebookresearch/dynalab/pull/78

I've run the instructions on a "Deep Learning AMI (Ubuntu 18.04) Version 44.0" AWS instance to check that everything works ok.