Potential issue w/ lifetime of RDMA async read buffers

[TylerADavis]

Currently the buffer that an asynchronous RDMA read reads into has its lifetime tied to the future returned by the operation. This has been working so far, but it seems possible that if the future returned by the function were to go out of scope before the read operation completed that buffer to read into would be deallocated, resulting in the read operation potentially overwriting memory in use elsewhere.

It may be beneficial to find a way of tying the lifetime of the receive buffer to the asynchronous operation (perhaps place a shared pointer inside of RDMAOpInfo).

[TylerADavis]

Another alternative would be to place a shared pointer inside of RDMAMem, but that doesn't seem very clean either.