stand: zfs: handle holes at the tail end correctly

[kevans]

Summary

Holes at the beginning and in the middle will get hole entries, but
holes at the tail end should be derived from the size recorded in the
dnode's attributes.

If we've walked past the end of the datablks, stat the dnode and zero
fill the buffer as long as we're not requesting past the end of the
file.

PR: 263407

[kevans]

Restrict this a little bit, to match dmu_read_impl(). This "tail hole" is only
handled by both illumos and openzfs if maxblkid == 0, which implies a weird
block size. Clip the read to just what's covered by the block, and zero out
the remainder like the kernel will. If we continue to hit blocks above maxblkid
for maxblkid > 0, there's still something weird happening.

[tsoome]

stand/libsa/zfs/zfsimpl.c

154

This part is now unrelated, perhaps would need either note or something?