Benchmarks are much better for opaque fills and slightly worse on non
opaque fills. I think that on balance, this is still a win.

When the source is uniform(color.RGBA{0x11, 0x22, 0x33, 0xff}):
name        old time/op  new time/op  delta
FillOver-8   966µs ± 1%    32µs ± 1%  -96.67%  (p=0.000 n=10+10)
FillSrc-8   32.4µs ± 1%  32.2µs ± 1%     ~      (p=0.053 n=9+10)

When the source is uniform(color.RGBA{0x11, 0x22, 0x33, 0x44}):
name        old time/op  new time/op  delta
FillOver-8   962µs ± 0%  1018µs ± 0%  +5.85%   (p=0.000 n=9+10)
FillSrc-8   32.2µs ± 1%  32.1µs ± 0%    ~     (p=0.148 n=10+10)

Change-Id: I52ec6d5fcd0fbc6710cef0e973a21ee7827c0dd9
Reviewed-on: https://go-review.googlesource.com/28790Reviewed-by: David Crawshaw <crawshaw@golang.org>