Use simple masking for AVX2 version of CopyAlpha so it can be implemented using a more generic bit mask function in future, and use more broadly known and optimized opcodes that will always be fast.  Same performance as vblend.
BUG=none
TEST=CopyAlpha*
R=johannkoenig@google.com

Review URL: https://webrtc-codereview.appspot.com/2393005

git-svn-id: http://libyuv.googlecode.com/svn/trunk@813 16f28f9a-4ce2-e073-06de-1de4eb20be90