Align SimdLib.h API docs with actual behavior for gradient, feature-difference, and alpha pipelines

[Copilot]

The issue was outdated/inexact descriptions in SimdLib.h for several image-processing APIs, especially around feature-difference math and alpha compositing semantics. This update rewrites those docs to match implementation-level behavior and parameter meaning.

• Abs gradient + feature difference semantics

  • Updated SimdAbsGradientSaturatedSum wording to explicitly describe saturated |dx| + |dy| behavior and border handling.
  • Corrected SimdAddFeatureDifference formula to the implementation’s actual excess term and saturating accumulation:
    • excess = max(max(value - hi, lo - value), 0)
    • difference = min(difference + ((weight * excess * excess) >> 16), 255)

• Alpha blending family clarified

  • Reworded SimdAlphaBlending, SimdAlphaBlending2x, SimdAlphaBlendingUniform, and SimdAlphaFilling around true per-channel equations.
  • Explicitly documented DivideBy255 rounding form used by implementation instead of plain /255 shorthand.

• BGRA → YUV420p alpha path documented precisely

  • Updated SimdAlphaBlendingBgraToYuv420p docs to reflect real behavior:
    • Y: per-pixel BGR→Y then alpha blend with destination Y.
    • U/V: 2x2-averaged BGR and averaged alpha, then blend with destination U/V.

• Premultiply / unpremultiply channel-layout behavior

  • Refined SimdAlphaPremultiply and SimdAlphaUnpremultiply descriptions to reflect:
    • alpha-channel preservation,
    • clamp behavior on unpremultiply,
    • argb-controlled alpha index (3 for BGRA/RGBA, 0 for ARGB).

// SimdAddFeatureDifference (documented to match implementation):
excess = max(max(value[i] - hi[i], lo[i] - value[i]), 0);
difference[i] = min(difference[i] + ((weight * excess * excess) >> 16), 255);

[Copilot]

Pull request overview

This PR updates the public C API documentation in SimdLib.h to accurately reflect the library’s implemented behavior for CRC, absolute-difference/gradient, feature-difference accumulation, and multiple alpha blending / alpha conversion pipelines.

Changes:

• Clarifies CRC-32 / CRC-32C parameters (polynomial form, init/final XOR semantics) and tightens wording.
• Aligns gradient and feature-difference documentation with implementation details (including saturation/clamping and exact formulas).
• Documents alpha blending math using the library’s DivideBy255 rounding form and clarifies BGRA→YUV420p alpha blending semantics and premultiply/unpremultiply channel layout behavior.