reduce latency in __purge_vmap_area_lazy

[Christoph Hellwig]

Hi all,

this is my spin at sorting out the long lock hold times in
__purge_vmap_area_lazy.  It is based on the patch from Joel sent this
week.  I don't have any good numbers for it, but it survived an
xfstests run on XFS which is a significant vmalloc user.