Prepare multi gpu support

[agnesLeroy]

closes: please link all relevant issues

PR content/description

This PR prepares the code base for multi-gpu support. At the moment all functions fall back to using only GPU 0, but we'll be able to add multi-GPU support little by little on top of this.

Check-list:

• [ ] Tests for the changes have been added (for bug fixes / features)
• [ ] Docs have been added / updated (for bug fixes / features)
• [ ] Relevant issues are marked as resolved/closed, related issues are linked in the description
• [ ] Check for breaking changes (including serialization changes) and add them to commit message following the conventional commit specification

[agnesLeroy]

@tmontaigu I've done what you suggested and created a MultiGpuFunctionExecutor for the tests, but actually I'm thinking a better way of doing this would be to pass a &[CudaStream] to all entry points, and they would make sure GPU 0 is only ever used when there is no multi-GPU support. Otherwise in the code as it is now, users could try to execute things on GPU 1, but that would conflict with the multi-GPU implementation of add, which supposes the keys, inputs & outputs are on GPU 0.

[tmontaigu]

If you think that is better you can do it

[mironov788]

Very interesting!!

[agnesLeroy]

@tmontaigu I gave a tour of the changes to @pdroalves earlier, so he has a good view of what's in the PR normally. I can go over it with you as well if you want, as this is a lot of changes...

[agnesLeroy]

@tmontaigu thanks a lot for your feedback! I pushed a separate commit where I took your comments into account.