Fix problems with new write_record() interface:

Fix problems with new write_record() interface:
- we were not incrementing the size properly
  - block > i_size >> blkbits was never true when it needed to be
  - new size was always 1 block too small
- when size was set, it didn't make it to disk, because
  ext3_get_block_handle() updated size after marking the inode dirty

Proper fix is to move i_disksize update into ext3_splice_branch(), so that
it is done before inode is marked dirty.  For now, we mark inode dirty
again in fsfilt, so we don't need a new kernel for Wed LLNL debug shot.

Make fsfilt_ext3 and fsfilt_extN identical again (cosmetic changes only).