Fix alignment check and stride usage in AbsDifference function

[yuchangminghit]

This pull request addresses two issues in the AbsDifference function:

1.Alignment check for output image c:
When the input images a and b are aligned, the output image c is directly assumed to be aligned without performing an alignment check.
In NEON, it assumes 16‑byte alignment:

uint8_t* _p = (uint8_t*)__builtin_assume_aligned(p, 16);
vst1q_u8(_p, a);

In AVX2, it assumes 32‑byte alignment:

_mm256_store_si256(p, a);

2.Stride assignment for output pointer:
Change c += bStride; to c += cStride; for clarity, although the original code does not affect the calculation result.