Remove alignment constraint for SSE2. Allows the optimized function to be used with unaligned memory, improving performance in that use case.  Hurts performance on core2 and prior where memory was faster with movdqa instruction.
BUG=365
TESTED=psnr, ssim and djb2 unittests pass.
R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/22859004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1100 16f28f9a-4ce2-e073-06de-1de4eb20be90