Sanitize bitmaps produced by libgif

Sanitize bitmaps produced by libgif

Motivation: fix the bug linked to below.

Some rare GIFs in the wild have a color table of size smaller than 256
and have one or more pixels with values that index out side of the
colortable.  Since it would be expensive to check the index every time
we do a color lookup, we sanitize it at the source, in
SkImageDecoder_libgif.cpp.  The new function sanitize_indexed_bitmap
checks each pixel to see if the index is outside of its allowed mask,
and if so, sets that pixel to point at color 0.

The bit mask optimization in this function relies on the fact that GIF
requires the color table size to be a power of 2.

To test:

	$ skia_images_gif_suppressDecoderWarnings=0 \
		..../render_pictures \
		--bbh grid 256 256 --clone 1 --config 8888 --mode tile 256 256 \
	    -r ..../http___www_cafe24_com_.skp -w /tmp

	This should show verbose errors.

BUG=skia:1946
R=robertphillips@google.com

Committed: https://code.google.com/p/skia/source/detail?r=12786

[hal.canary, 2013-12-19 20:16:41]

PTAL

[robertphillips, 2013-12-19 20:30:38]

lgtm + comment request

https://codereview.chromium.org/103343006/diff/1/src/images/SkImageDecoder_li...
File src/images/SkImageDecoder_libgif.cpp (right):

https://codereview.chromium.org/103343006/diff/1/src/images/SkImageDecoder_li...
src/images/SkImageDecoder_libgif.cpp:188: static void
sanitize_indexed_bitmap(SkBitmap* bm) {
// Indexed GIFs with fewer then 256 color entries will sometimes index out of
bounds

[hal.canary, 2013-12-19 20:50:11]

https://codereview.chromium.org/103343006/diff/1/src/images/SkImageDecoder_li...
File src/images/SkImageDecoder_libgif.cpp (right):

https://codereview.chromium.org/103343006/diff/1/src/images/SkImageDecoder_li...
src/images/SkImageDecoder_libgif.cpp:188: static void
sanitize_indexed_bitmap(SkBitmap* bm) {
On 2013/12/19 20:30:38, robertphillips wrote:
> // Indexed GIFs with fewer then 256 color entries will sometimes index out of
> bounds

Done.

[hal.canary, 2013-12-19 20:57:05]

I also tested this twice with the new Cluster Telemetry Skia Tryserver, and
found that my before images change and my after images don't.  Also, they look
better - no random colors.

[hal.canary, 2013-12-19 20:57:50]

Committed patchset #2 manually as r12786 (presubmit successful).