Guard against accidental io.EOFs

[puellanivis]

There are a number of situations where we should really be able to know that the code path doesn’t contain an io.EOF. As a precaution, we put a guard statement in each of these paths that should prevent an unintentional io.EOF from falling through.

Instead, we return ErrSSHFxConnectionLost instead. I’m not sure this is precisely the correct return value we want to use instead of an io.EOF, but I cannot think of what other kind of an error we should return.

[puellanivis]

Integration test failure appeared to have been caused by a race condition between two StatVFS commands returning slightly different values.

[puellanivis]

This not only didn’t solve the issue we were facing, but it’s also obsoleted by the other refact, so I’m going to close this PR.