refactor: unify ws_mask/ws_unmask via xor_impl function pointer

Replace duplicated SIMD bodies in ws_mask and ws_unmask with three static
implementations (xor_scalar, xor_sse2, xor_avx2) sharing a common signature
(src, dst, mask_ptr, mask32, length). A single xor_impl function pointer,
set once at the end of _init_cpu_features, eliminates all per-call branches
and the avx2_available global. ws_unmask passes buf as both src and dst for
in-place XOR; ws_mask passes src and out+offset.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: mvogttech