Everyone makes mistakes, so it also happens that a crate author discovers that they made a mistake, and then re-release.
Thing is: when the queue is very full, this can lead to multiple versions of the same crate being in the queue:
What would be cool in this case is that we deprioritize the older releases (to the same priority as crates-io-triggered rebuilds, to 5, if the prio is not already lower).
I think even when the older releases are yanked, we still might want to have the docs, so I lean towards still building the docs ( we could deprioritize yanked releases even more).