if I didn't import enough watch-only addresses] and then use listunspent RPC.
So import a million, performance is acceptable with very large numbers of keys.
This sounds like you're asking about something for yourself to run.
Another issue is that currently pruning disables the Bitcoin Core wallet so
Mostly as a precaution as it hadn't been adequately tested. It's enabled in master now and has been for a long time; considering the crazy things you were considering (concurrent access to the wallet database?!?) running master should be a no brainer.
If it would be useful to you, supporting rescan that goes back as far as the non-pruned data shouldn't be a big deal-- mostly an interface question of changing the rescan argument to take a depth and permit the import w/ rescan when the depth is compatible with the current level of pruning. For what I think you're trying to do you really only need to rescan back a block or two, no?