Function convert_bgr_to_argb_avx2 

pub unsafe fn convert_bgr_to_argb_avx2(
    dst: *mut u32,
    src: *const u8,
    count: usize,
)

BGR24 => ARGB32 conversion using AVX2 vpshufb (8 pixels/iter)

Load exactly 24 bytes (8 BGR pixels) without buffer overflow by constructing the YMM from two separately loaded 128-bit lanes. vpshufb operates per 128-bit lane independently, so each lane must have its 4 BGR pixels in the first 12 bytes, followed by zeros.