Request for fuselets

[anyzelman]

A request came in for accelerating process-local parts of existing HPC solvers using the nonblocking backend. This should be easy to do, and this issue proposes to demo an easily-extensible set of basic level-1, level-2, and mixed level-{1,2} sequences that can be fused. We will call elements of these sets a fuselet, and expose the set as part of the transition path as fuselets.