feat: Support JPEG decode directly to YUV planes

Hello,

It would be very useful if Wuffs could support decoding JPEGs directly to YUV planes (YCbCr 4:2:0 / 4:2:2 / 4:4:4) instead of always converting to RGB/RGBA on the CPU.

This would help applications that:

upload decoded images directly to the GPU,
perform color conversion in shaders,
or use YUV textures directly.

Avoiding the CPU-side YUV → RGB conversion could reduce:

CPU usage,
memory bandwidth,
temporary allocations,
and GPU upload size on bandwidth-constrained devices.

This is especially useful for embedded systems, ARM devices, smart TVs, and low-end/mobile GPUs. TurboJPEG/libjpeg already support YUV-domain decode paths, so having similar support in Wuffs would be very valuable.