Feature suggestion: distributed learning on volunteer compute

Another feature idea is for a case where we allow the user to upload their model, which we will train for them and evaluate on a chosen benchmark with a chose metric.

We can think of ways to use open-source distributed-learning libraries like [this one](https://github.com/learning-at-home/hivemind) for **very** distributed training (taking into account a possibility where compute is not guaranteed for the whole training process and coincidently) on volunteer computers.

And there are [known success stories] of volunteer compute being used for costly scientific computations (https://en.wikipedia.org/wiki/LHC@home#:~:text=The%20project's%20computing%20power%20isand%20other%20experimental%20particle%20accelerators.&text=The%20project%20is%20run%20withteraFLOPS%20as%20of%20March%202023.)

I understand that this is very far-fetched but hopefully Brain-Score team grows enourmously one day, and there will be enough people to handle this as well!