remove sse2 functions that also have ssse3

Remove sse2 functions that also have ssse3

ARGBBlendRow_SSE2, ARGBAttenuateRow_SSE2, and MirrorRow_SSE2
Since vast majority of CPUs have SSSE3 now, removing the SSE2
improves the performance of CPU dispatching.

R=harryjin@google.com
BUG=none

Committed: https://chromium.googlesource.com/libyuv/libyuv/+/2cc1a2b233488297d2590f744fdee8ff604f5c11

[harryjin, 2015-09-30 02:42:08]

https://codereview.chromium.org/1377053003/diff/20001/include/libyuv/row.h
File include/libyuv/row.h (left):

https://codereview.chromium.org/1377053003/diff/20001/include/libyuv/row.h#ol...
include/libyuv/row.h:124: #define HAS_MIRRORROW_SSE2
So for the very few SSE2 only case, things just got a little slower but nothing
breaks right?

[harryjin, 2015-09-30 17:42:59]

lgtm

[fbarchard, 2015-09-30 21:23:43]

Tested on Prescott emulator.. passes.
Performance as measured on Haswell:
LIBYUV_FLAGS=-1 LIBYUV_DISABLE_ASM=0 LIBYUV_DISABLE_SSSE3=1
LIBYUV_DISABLE_AVX2=1 LIBYUV_WIDTH=1280 LIBYUV_HEIGHT=720 LIBYUV_REPEAT=9999
out/Release/libyuv_unittest --gtest_filter=*ARGBBlend_Opt

AVX2
ARGBAttenuate_Opt (2716 ms)

SSSE3
I420Mirror_Opt (959 ms)
ARGBAttenuate_Opt (3538 ms)
ARGBBlend_Opt (4411 ms)

SSE2
I420Mirror_Opt (1289 ms)
ARGBAttenuate_Opt (4408 ms)
ARGBBlend_Opt (4815 ms)

C
I420Mirror_Opt (4371 ms)
ARGBAttenuate_Opt (18547 ms)
ARGBBlend_Opt (15572 ms)

Impact is the SSE2 version goes away and falls back on C.  Depending on
platform, there are still CPUs out there with SSE2 only, but they are likely not
running video hangouts.  Prefer to invest in AVX2 versions or these functions.

[fbarchard1, 2015-09-30 21:24:49]

Committed patchset #2 (id:20001) manually as
2cc1a2b233488297d2590f744fdee8ff604f5c11 (presubmit successful).