MRCuda: rename RuntimeInfo to DeviceInfo; report GPU name and total memory

[Fedr]

Summary

Extends the information that MRCuda reports about the local GPU, and renames the API to better reflect that it describes the device rather than just runtime versions.

Changes

• Renamed Cuda::RuntimeInfo → Cuda::DeviceInfo and Cuda::getRuntimeInfo() → Cuda::getDeviceInfo().
• Added two fields to the struct:
  • size_t totalGlobalMem — total global memory on the device in bytes (not all of it is available to our process).
  • std::string name — device name.
• These fields plus the compute capability are now read from a single cudaGetDeviceProperties() call, replacing the separate cudaDeviceGetAttribute() calls.
• Clarified the doc comments: driverVersion is the maximum CUDA version supported by the driver, runtimeVersion is the application's CUDA version.
• Updated callers (isCudaAvailable, MRTestCuda) to the renamed function.