Integrate flows with parallel ODE problem

[SouthEndMusic]

This is an idea that came up after observing that #1431 is very bad for performance, see also #1444 (comment). So my thought was:

What if set up a parallel ODE problem next to the main one, where the du is given by the flows that come from the main solver? For performance and lack of information at intermediate timesteps it should use a relatively simple explicit algorithm. I would like to hear your thoughts on this regarding code complexity etc.

[visr]

This can probably be closed if #1468 works out well.