Match the changes made to the downstream API on the NRM side.

This patch cleans up the code, use better names for the headers, remove
the last bits of C++ and add autotools to compile the library.

This patch also splits the PMPI library into a separate .so, to make it
easier to only deal with the core downstream API.

Instead of transmitting clock times, the downstream API is changed to
allow application side aggregation of phases below a damper value. This
is to decrease the frequency of messages sent.

The downstream API will now transmit the actual time a CPU spent in
computation and the total phase time.

See Issue #2