How to apply the usual care measures?

[dev8765]

How to apply defrag (HDDs), trim (SSDs) and scrub & dedup (ZFS)?
Must these be applied to the raw devices (e.g. /dev/sdb), to the NonRAID block devices (/dev/nmd1p1), to the ZFS-pools (disk1), to the mounts (/mnt/diskN) or to the virtual filesystem created by mergerfs?
Should the NonRAID array be stopped beforehand, presumably followed by nmdctl check?
Any other tips?

[qvr]

You should apply all filesystem specific maintenance to the nonraid block devices. You should not modify the raw devices ever, otherwise the array parity will become invalid.

On Ubuntu at least, any zfs pools created on the nonraid disks get automatically scrubbed monthly by the standard ZoL scrub schedule.

The driver does not support TRIM (which is why SSDs are not recommended in the array).

[dev8765]

SSD-arrays are possible in general, write speeds keep quite consistent over time, performance degradation caused by missing TRIM is not that great or even noticeable
Additional it should be possible to manually stop the array, trim all the drives, recalculate parity and restart the array

[qvr]

Yes, if you are willing to accept the risk of the array disks not being parity protected while parity is being recalculated, then that will work.