Preserve alpha in raw AVIF and HEIF encoding

[gregbenz]

What changed

Preserve alpha when AVIF or HEIF is encoded from a supported raw RGBA input.

• API-0 takes alpha from the original HDR input (RGBA1010102 or RGBAHalfFloat).
• API-1 takes alpha from the supplied RGBA8888 SDR base image.
• Non-opaque alpha is attached to the encoded base as a straight 8-bit alpha plane.

Previously, the RGB-to-YCbCr conversion discarded alpha before encoding, so accepted RGBA inputs always decoded as fully opaque. For API-0, alpha must come from the original HDR input because the internally tone-mapped SDR image has already been made opaque.

Half-float alpha is quantized to 8-bit, matching the current 8-bit base-image encoding. The public C API and non-RGBA paths are unchanged.

Testing

Added public-API regression coverage for both AVIF and HEIF:

• API-0 with RGBA1010102
• API-0 with RGBAHalfFloat
• API-1 with different HDR and SDR alpha, confirming the SDR base is the source
• intermediate alpha values (0, 85, 170, and 255)
• direct verification that the encoded base carries non-premultiplied alpha

All six cases fail on current main because decoded alpha is always 255; they pass with this change. The full unit suite also passes: 1,073 passed, with 224 existing parameter skips.

Compressed-input/API-4 AVIF/HEIF transcoding is outside this focused fix and is tracked separately in #454.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.