xfs: refactor and improve inode iteration

This next series refactors all the inode walking code in XFS into a
single set of helper functions.  The goal is to separate the mechanics
of iterating a subset of inode in the filesystem from bulkstat.

First we clean up a few weird things in XFS, then build a generic inode
iteration function.  Next, we convert the bulkstat ioctl to use it, then
fix a few things from some of the code we saved from the old bulkstat
inode iteration code.  After that, we restructure the code slightly to
support the inumbers functionality, and then port the inumbers ioctl to
it too.

Finally, we introduce a parallel inode walk feature to speed up
quotacheck on large filesystems.  The justification for this part is a
little questionable since it needs further discovery of what hardware
and software this works best on.  It's also an open question of whether
or not bulkstat could be optimized further.