Change memmove's back into memcpy's. No functional change. Not flag protected.

Change memmove's back into memcpy's. No functional change. Not flag protected.

memmove is presently ~2.5x faster than memcpy, however isn't inlined with
-minline-all-stringops. It turns out that when these effects combine, the net
result is a no-op for performance.

So let's switch back to memcpy, because memcpy is likely to be improved in the
future, and seeing memmove's where they aren't needed is somewhat surprising.

Merge internal change: 128366346

BUG=

[Ryan Hamilton, 2016-07-27 21:32:31]

rch@chromium.org changed reviewers:
+ zhongyi@chromium.org

[Zhongyi Shi, 2016-07-27 21:37:00]

lgtm