Add GET_STACK_USAGE() for arm64.

[markj]

Summary

I noticed while testing builds on a ThunderX that the GEOM
service threads were quite busy. Turns out that arm64 was
missing a macro that's required in order to enable so-called
direct dispatch in GEOM. Without that, all I/O passes through
these threads.

The implementation is just copied from other arches. I will
look into unifying them, but this works for now.

Test Plan

buildkernel, verified that the GEOM threads are no longer busy