Research moving reporting IO to a dedicated thread

[RobertJacobsonCDC]

It is easy to accidentally create a bottleneck with logging or reporting IO. We should experiment with moving IO to a separate thread from simulation computation. This needs to be done carefully, because we don't want to interfere with event order or determinism of the reported data. It's not clear:

• if it's possible to do this while maintaining our determinism guarantees;
• if doing this gives a meaningful net benefit for IO heavy simulations.

This might be low-priority, although Guido did report an unexpected runtime cost in his epi-isolation experiment, so maybe not?