- Huge refactoring: functions fit in a new library, libcapulet
- Rewriting of the `rdma_standalone` example with new library
- Separate MR registration from other RDMA operations
- Allow passing custom access flags to QP

- Remove unused `udp_send` and `udp_recv`
- Add an example file to send
- Organize files into directories

- Allow passing server name by CLI
- Read server content from STDIN
- Add documentation for launching the demo

- Use of CQ of length 2: Send + Recv
- Allow node to set its own LID

- Initialize server with Recv requests
- Reduce RX pool size to match other sizes
- Update access flags
- Pool CQ for completion of Recv requests

While the RDMA server and client are very similar, two files are currently used
for an development purposes. The client acts as an UDP client and will try to
read available memory from it's peer.

Add a Global IDentifier to the UDP packet in order to be able to communicate with a distant machine.

Last step before introducing a client is to allow server to connect to the client QP.
That is done when switching the state of the local QP from Init to RTR.

Parameters need to be exchanged in both way using UDP, so the initial (unused) `capabilities` packet
was dropped in favor of sending client's parameters with the request.

Merge together the UDP exchange example with the server side of the
RDMA example.

Add an example file which creates an RDMA device, allocates a Memory Region
and prepares a Queue Pair for a client to read the MR.

Create two functions which can be called from the outside to exchange RDMA informations using UDP.
The example files now refer to these functions.

- Use a bin/ directory for output
- Define `make all` as the default recipe

Before starting a RDMA communication between two nodes, local
informations must be exchanged. The proposed example uses UDP
for that purpose.