run model submissions in parallel

As the number of benchmarks to evaluate models on grows, we should consider running models in parallel instead of the current serial execution(see jenkins configuration and [here](https://github.com/brain-score/brain-score/blob/14b7d3d3540c007df041c46e5211c25dd55e0fa7/brainscore/submission/evaluation.py#L103)). This would involve launching and coordinating multiple jobs to aggregate the results in an email to the user. Some jobs we might also want to run on GPU instead of CPU, ImageNet-C in particular takes too long to run on CPU.

This is a pretty big task since the entire submission logic is _inside_ the [python](https://github.com/brain-score/brain-score/blob/14b7d3d3540c007df041c46e5211c25dd55e0fa7/brainscore/submission/evaluation.py#L103) call that is run on OpenMind, i.e. _after_ the job is created. We would thus need to re-write that logic and e.g. spawn OpenMind jobs from the python script.