Support native bit operations on Windows

[aabbdev]

Summary

• replace GCC/Clang-only 64-bit count builtins with exact C++17 helpers
• retain compiler builtins on GCC/Clang and use portable bit operations under MSVC
• unblock Windows x64 wheels for rosa-torch-native 0.3.0

Validation

• native wheel rebuilds on macOS arm64
• top-1 and RLBWT native smoke tests pass
• Windows failure reproduced at __builtin_clzll, __builtin_ctzll, and __builtin_popcountll
• cross-platform cibuildwheel dry run will execute on this branch before merge