xfs: a couple of realtime growfs fixes

While I was taking a closer look at Chandan's earlier fixes for the
realtime growfs code, I realized that fstests doesn't actually have a
test case for growing a realtime volume.  I wrote one for testing rmap
expansions on the data device and kludged it to work for realtime, and
watched the kernel trip over a ton of assertions and fail xfs_scrub.

The two patches in this series fix the two problems that I found.  The
first is that inode inactivation will truncate the new bitmap blocks if
we don't set the VFS inode size; and the second is that we don't update
the secondary superblocks with the new rt geometry.

I'll cough up a test case after this patch.

v2: fix some review comments, add a locking annotation fix