mpl/cuda: Add device id to handle conversion API

The libfabric memory registration API requires a CUdevice handle as
input. Add MPL_gpu_device_id_to_handle to support this
functionality. HIP and ZE do not need to implement this API since it
is only used within MPL_HAVE_CUDA preprocessor blocks.

src/mpl/include/mpl_gpu_cuda.h

@@ -13,6 +13,7 @@ typedef cudaIpcMemHandle_t MPL_gpu_ipc_mem_handle_t;
 typedef struct cudaPointerAttributes MPL_gpu_device_attr;
 typedef int MPL_gpu_request;
 typedef cudaStream_t MPL_gpu_stream_t;
+typedef CUdevice MPL_gpu_device_handle_t;
 
 /* Note: event variable need be allocated on a gpu registered host buffer for it to work */
 typedef volatile int MPL_gpu_event_t;
@@ -22,4 +23,6 @@ typedef volatile int MPL_gpu_event_t;
 
 #define MPL_GPU_DEV_AFFINITY_ENV "CUDA_VISIBLE_DEVICES"
 
+#define MPL_gpu_device_id_to_handle(...) cuDeviceGet(__VA_ARGS__)
+
 #endif /* ifndef MPL_GPU_CUDA_H_INCLUDED */