Updated request-generating RMA operations to correctly handle
communication with MPI_PROC_NULL.

Reviewer: goodell

Communication operations on shared memory windows now perform the op directly
on the shared buffer.  This requried the addition of a per-window interprocess
mutex to ensure that atomics and accumulates are performed atomically.

Reviewer: buntinas

This patch more efficiently handles request-generating operations that target
the local process and MPI_PROC_NULL.  These special cases return completed
generalized requests, which cost less overhead.  This also makes space for a
more efficient implementation when direct, shared memory operations are
implemented.

Reviewer: goodell

Note: Request-generating operations can only be called within a passive target
epoch.

Reviewer: goodell

When called with a target process of PROC_NULL, RMA request-generating
operations did not return a valid request handle.

Reviewer: goodell

This patch add a Grequest wait function, is needed to support waitall with RMA
ops.

Reviewer: buntinas

This adds a simple implementation for request-based RMA operations.  The
implementation uses generalized requests as follows:

 1. Insert operation into the RMA ops list
 2. Create a generalized request
 3. When the request is polled, flush the window and mark completed

Reviewer: buntinas