Initial fuzzing harness for animated images

[Shnatsel]

Animated images with the "loop forever" flag currently cause an infinite loop in the decoder.

The AFL seed images didn't seem to contain any animated ones, so I made these seed images myself from the sample image linked from the official Google documentation: webp_animated_fuzz_seeds.tar.gz

Resolves (?) #31

[fintelia]

At the moment, next_frame intentionally starts over back at the beginning of the animation until it has concluded loop_count many passes. With the "loop forever" flag it'll return each frame an infinite number of times.

For displaying the image, that seems like desirable behavior. However, if you're trying to collect all the frames of the animation (say to convert them to a different format) then it is far less helpful. Perhaps the right approach would be to expose extra methods to query the loop count and number of frames?

[Shnatsel]

I think there needs to be a lower-level interface that exposes all the frames and separately the data on how many times they should be looped, and optionally a higher-level one that does the right thing for displaying the image (that's how the current interface works).

I wonder, does the current interface decode all the frames once and store them in memory, or does it decode them over and over?

[fintelia]

The current interface decodes each frame as it is requested and also does compositing at the same time. Which can involve blending pixels with the values from previous frames. And the raw frame data can also be limited to only the "cropped" section that changed from the previous frame so we account for that as well.

Other than not automatically looping the animation, are you imagining there would be any other differences with the lower-level interface?

[Shnatsel]

Well, an image editing tool like GIMP would require access to the raw data of the layers - but I don't think anyone else would care. I think just not automatically looping is a good start. We can add the even lower-level interface useful for image editors once someone who wants to use it actually shows up. I feel it's easy to design something useless in absence of actual users who can tell us what they need.

[fintelia]

I'll go ahead and merge this, and then API improvements can come as a followup.