My PCs are too old for NVMe, so I can't really help. But I run NetBSD on a M.2 (SATA) SSD which uses a 512B logical sector size, with an underlying 4K physical arrangement, as returned by atactl identify. To minimize the translation overhead, all partitions are aligned at 4K boundaries, and
formatted as FFSv2(ea) with a 4K fragment size and a 4K ยท 8 block size.
In the example below, providing that wedge dk1 is aligned to a physical sector boundary, you can format the partition with a fragment size matching the the real sector arrangement, by issuing:
$ newfs -O 2ea -V2 -b 32768 -f 4096 /dev/rdk1
