Refine SimdLib API docs for BGRA/BGR conversion functions

[Copilot]

The SimdLib.h descriptions for several BGRA/BGR conversion APIs were too generic and, in places, inaccurate about subsampling, alpha handling, and dimensional constraints. This update aligns those comments with the actual implementation behavior.

• YUV conversion semantics (BGRA → YUV*)

  • Clarified per-format sampling behavior for:
    • SimdBgraToYuv420pV2 (4:2:0, 2x2 chroma averaging, even width/height)
    • SimdBgraToYuv422pV2 (4:2:2, horizontal-pair chroma averaging, even width)
    • SimdBgraToYuv444pV2 (4:4:4, per-pixel chroma)
    • SimdBgraToYuva420pV2 (4:2:0 YUV + full-resolution alpha copy)
  • Explicitly documented that BGRA alpha is ignored for YUV-only outputs and copied for YUVA output.

• Bayer/BGRA conversion behavior

  • SimdBgrToBayer: documented mosaic selection semantics from bayerFormat and even-dimension requirement.
  • SimdBgrToBgra: documented constant-alpha fill behavior.
  • SimdBgr48pToBgra32: documented 16-bit planar-to-8-bit channel extraction behavior (endianness-dependent byte choice) and alpha fill.

• Color-space output details

  • SimdBgrToGray: documented weighted grayscale conversion formula with rounding.
  • SimdBgrToHsl / SimdBgrToHsv: documented channel layout ([H,S,L], [H,S,V]) and 8-bit component ranges.

// SimdBgraToYuv420pV2 docs now state:
// - Y: per-pixel from B,G,R
// - U/V: one sample per 2x2 block from averaged B,G,R
// - constraints: width and height must be even