ZFS vn_rele_async: ctach up with the use of atomics for the vnode use count

[avg]

Summary

It's not sufficient nor required to use the vnode interlock when checking
if we are going to drop the last use count as the code in vputx() uses
atomic operations for both checking and decrementing the use code.
Apply the same method to vn_rele_async().
While here, remove vn_rele_inactive(), a wrapper around vrele() that didn't
add any value.

Also, the change required making vfs_refcount_release_if_not_last() public.
I've made vfs_refcount_acquire_if_not_zero() public as well.
They are in sys/refcount.h now. Maybe this is not the best header or maybe
the names need to drop "vfs" prefix.

Sponsored by Panzura.

[kib]

sys/sys/refcount.h

83 ↗

(On Diff #40802)

The symbol from refcount.h should not abuse vfs_ namespace.

Also, note that acquire in the name of the function does not follow common use of the term WRT the atomic API.

[avg]

sys/sys/refcount.h

83 ↗

(On Diff #40802)

I can drop the prefix if that's the request and Mateusz does not have a different opinion.

Regarding "acquire" -- yes, it's meaning is different from atomics, but is the same as in refcount_acquire.

[mjg]

sys/sys/refcount.h

83 ↗

(On Diff #40802)

These inlines were supposed to be temporary bandaids until a general refcount_* primitives get implemented. I don't remember now what exactly stopped the effort at the time, someone(tm) should revisit this. I have a weak recollection that part of the point here was to either drop OR grab a lock while within the primitive. I guess a weaker variant which is equivalent to what's seen here is good enough for general use as well.

In the meantime I don't see a problem with having these named like this. The intent is to remove them soon anyway and just drop the vfs_ prefix here. Renaming is completely indifferent to me.

[avg]

drop vfs_ prefix from the refcount "if_not" functions