Create error handler for remotes

[pared]

As in
#2866 (comment)

@MrOutis @Suor @efiop
I would like to ask your opinions on that, though AFAIK, @efiop agrees.

[efiop]

So the change would be that each _download/_upload knows which specific errors to catch (e.g. ClientError for s3) and the rest falls through and is either being caught in RemoteBASE.download/upload if it is something common but remote-related, or it is caught by main.py (e.g. ulimit issue). Currently we catch everything in RemoteBASE.download/upload and treat it as something recoverable, which is not the case.

[Suor]

I don't see much value in this refactoring unless we actually do something useful with that specific errors, i.e. retry or dynamic_chunk_size. Otherwise we will just add lots of code, which may make it harder for us to actually do something useful later.

[ghost]

Can't imagine how would it look like, @pared, but I agree that we should treat exceptions better

[efiop]

Migrating to fsspec that adapts cloud specific errors to OSError's and even retries stuff if needed. Closing as stale.