Optimise bitwise extraction and insertion

[lijunbie]

Summary

• Add guarded BMI2 fast paths for insertBits() and getValueOfBits() using _pdep_u64 / _pext_u64 on x86-64 builds compiled with __BMI2__.
• Keep the existing portable loop fallback for non-BMI2 builds, CUDA/HIP compilation, small fixed-size cases, and getValueOfBits() calls whose bit indices are not in increasing order.
• Add unit coverage for insertBits() and getValueOfBits() over empty, low-bit, multi-bit, and non-sorted extraction cases.

Closes #717

Validation

• Ran git diff --check.
• Ran an independent 64-bit equivalence check over 261 sample cases comparing the BMI2-style algorithms against the existing loop semantics.
• I could not run the full Catch2/CMake test suite locally because cmake is not installed in this environment.

[TysonRayJones]

That's some truly low effort AI slop