Make sibling twiddle lookup branchless

[falseywinchnet]

Motivation

• Prevent the compiler from lowering scalar twiddle selection into large conditional decision trees by making sibling twiddle lookup data-driven and branchless while preserving the mapping 0 -> 0, 1 -> 1, and m > 1 -> m ^ 1.

Description

• Replaced s_twiddle and sf_twiddle in src/detail/bruun_kernel.hpp with a branchless selector that computes u = static_cast<unsigned>(m), flip = static_cast<unsigned>(-(m > 1)) & 1u, and returns C[u ^ flip] / CF[u ^ flip] so sibling lookups become index operations instead of explicit m == 0 / m == 1 branches.

Testing

• Ran make test asm-check; build/tests/correctness and build/tests/api_c completed successfully and the generated SSE2/AVX2 assembly shows twiddle lookup lowering to setg/xor/load patterns instead of helper-level conditional branches.

---

Codex Task