Add CUDA bitset matrix support in pe_synth.h and pe_synth_cuda_u64_cones.cu

- Introduced CUDA helper functions for managing a row-major u64 matrix, enhancing performance for masked popcount and any checks in QM/Espresso scoring.
- Implemented RAII-style management for CUDA bitset matrices to ensure proper resource handling.
- Added CUDA kernel functions for row-wise popcount and any checks, optimizing bitset operations on the device.
- Updated existing logic to utilize the new CUDA capabilities, improving synthesis efficiency and flexibility.

Author: MacroModel