Size up IPC buffers to multiples of 2MB

Avoid potential CUDA error in concurrent communicator initialization


[ROCm/rccl commit: c8c68fb5f7]

projects/rccl/src/include/comm.h

@@ -23,7 +23,7 @@ struct cudaLaunchParams {
 
 #define CACHE_LINE_SIZE 128
 #define MEM_ALIGN 4096
-#define CUDA_IPC_MIN 2097152UL /* 2MiB - not currently used */
+#define CUDA_IPC_MIN 2097152UL
 
 struct ncclSendMem {
   union {