SSE2 accelerated opaque blitRect

Trying to make progress on a task we'd talked about and then parked months ago. (The blurs look less than feasible, although we could look at what's going wrong there.)

The chrome_scrollGmail benchmark was originally developed for this. The patch produces a 7% overall speedup; dividing the benchmark into narrow (< 31 px) and wide cases, it yields 15% on narrow rects, 5% on wide.

[TomH, 2012-03-16 20:32:20]

This version speeds up:
 20%:  path_fill_big_rect, rects_1, rects_1_stroke_4
 10%:  rects_3_stroke_4

[TomH, 2012-03-16 21:28:20]

Got rid of new headers, merged into SkBlitRow w/ a ColorRect32() sitting
alongside Color32(). Similar performance improvement.

TODO: new benchmarks, sprite version

[TomH, 2012-03-19 13:50:57]

Committed in r3423.