status update: worked through ms support, verified that win wouldn't have caused this corruption as the detach/attach of the lun wouldn't have caused the partition data to be touched at all.
also found a known bios bug that affected these machines and would cause this corruption. we had already patched our bios to a version that included the fix, so it seems there was a regression. still working with hw vendor...
I fixed the corruption by cheating. I used a hex editor to overwrite the corrupt 20k from a staging copy of the db. The affected pages hadn't changed, and a checkdb afterwards came up clear, so I feel pretty good about the fix. it it hadn't worked, restoring full plus diff would have been the next step. Not the recommend approach, but effective.