LU-1666 obdclass: reduce lock contention on coh_page_guard

LU-1666 obdclass: reduce lock contention on coh_page_guard

Define a per-page spinlock to get and put a cl_page instead of
grabbing per-object lock coh_page_guard.

Signed-off-by: Jinshan Xiong <jinshan.xiong@intel.com>
Change-Id: Iecf5e6840ab1a28edcf2c4bcde6a72c2f9b5bdae
Reviewed-on: http://review.whamcloud.com/3627
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: James Simmons <uja.ornl@gmail.com>
Reviewed-by: Ned Bass <bass6@llnl.gov>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Tested-by: Hudson
Tested-by: Maloo <whamcloud.maloo@gmail.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>