Modify communicator reiceve to write directly into user owned buffers

[hgangwar]

Adds a receive API that writes directly into user-provided storage, avoiding an intermediate allocation and copy for global communication.

The existing vector-returning Recv() API remains available for backward compatibility.

Usage

std::vector<double> values(size);

comm.Recv( values.data(), values.size(),
  redev::Mode::Synchronous);

Changes

• Added Recv(T* destination, std::size_t size, Mode).
• Added forwarding support in BidirectionalComm.
• Updated test_global_comm testcase.