Pool import slow after large amount of data written

Into 13.0-RC1 I've also just merged https://github.com/openzfs/zfs/pull/13022 , that should reduce import time by not scrubbing data of the last 2 committed TXGs.  The metadata are still scrubbed, since some people wish to have that as safety belt against pool corruptions.  For that reason it won't help much pools with dedup, since dedup table is a metadata, but should save few seconds for non-dedup pools.

I've asked Ryan to include the above PR into OpenZFS 2.1.2 update he is working on now for the 13.0-BETA1.  I'll keep this ticket for a while longer to investigate ZIL replay time.

I expect this patch significantly improve log spacemap load time: https://github.com/openzfs/zfs/pull/12789 .  As next step I am going to take a look on a ZIL replay time.

Profile looks alike to what I have reproduced on my system after trashing pool heavily to 73% fragmentation and reset the system under load to leave pool dirty.  It seems pool import takes a lot of time to time to reload space map log, doing that in single thread.