harden "get_bootloader_devices" function in upgrade scripts

[prakashsurya]

This issue is to follow up on the feedback on the get_bootloader_devices function used in the upgrade scripts. Specificially, the following comment:

Unless we test this really well (and have some automation to do this, which we don't), I'd suggest we simply fail if there is some funky configurations going on. If there is replacing going on, I feel like it would be much safer to just wait for the operation to finish before attempting upgrade. Do we have any production cases where we have a mirrored rpool, or is this just temporary for increasing the size of rpool? If that's so then we should also wait for the pool to go back to have a single device.

Originally posted by @pzakha in #210