[c10d] Add NCCL memory allocator (#145675)

This PR implements a small UI improvement over #133603.

It prepares a NCCL memory allocator in torch cpp and then pybind's it out, so that user can directly use it.

UI:
```
pool = torch.cuda.MemPool(backend.mem_allocator)
with torch.cuda.use_mem_pool(pool):
    tensor = torch.arange(1024 * 1024 * 2, device=device)
```

Pull Request resolved: https://github.com/pytorch/pytorch/pull/145675
Approved by: https://github.com/syed-ahmed, https://github.com/wconstab

torch/csrc/distributed/c10d/NCCLUtils.hpp

@@ -86,6 +86,10 @@ static_assert(
 #define NCCL_HAS_COMM_REGISTER
 #endif
 
+#if NCCL_VERSION_CODE >= NCCL_VERSION(2, 19, 0)
+#define NCCL_HAS_MEM_ALLOC
+#endif
+
 // Macro to throw on a non-successful NCCL return value.
 #define C10D_NCCL_CHECK(cmd, failureReason)                                   \
   do {                                                                        \