b=5654

b=5654
r=adilger

The last little bits of cleanup for bug 5654 (from Andreas's review):
- save_valid isn't needed in ll_setattr_raw, because either the function exits
  early or ia_valid gets cleared regardless
- protect i_size updates in ll_nopage and ll_ap_refresh_count
- change ll_page_matches to only match outgoing locks for real reads;
  readaheads no longer match locks which are in the process of being
  cancelled anyways