It can take a long time to run log recovery operation because it is
single threaded and is bound by read latency. We can find that it took
most of the time to wait for the read IO to occur, so if one object
readahead is introduced to log recovery, it will obviously reduce the
log recovery time.

Log recovery time stat:

          w/o this patch        w/ this patch

real:        0m15.023s             0m7.802s
user:        0m0.001s              0m0.001s
sys:         0m0.246s              0m0.107s
Signed-off-by: NZhi Yong Wu <wuzhy@linux.vnet.ibm.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>