Fix three problems related to lock extent revocation:

Fix three problems related to lock extent revocation:
- we were incorrectly calculating the "end" index value to be at the beginning
  of the next stripe instead of at the end of the current stripe (extra +1 for
  end, instead of using <= end for page indices)
- when comparing end << PAGE_CACHE_SHIFT for overflow against extent->end we
  used an unsigned long (which overflowed during the test ;-) so we always
  evicted pages to EOF (may have been a culprit in bug 2106 also)
- when matching against other locks (to see if page is under 2 locks) we
  were doing so in the file offset space (from page->index) instead of in
  the osc/extent space so we were matching against the wrong locks, if any

Change LL_CDEBUG_PAGE() to take a debug mask, so we can use it for error cases.
b=2765