add multithreaded source reading

[rinusser]

Currently the (only) scheduler waits for update intervals to pass, then reads each source in sequence.

Change this to a multithreaded model: each source should be read in a separate thread, so slow sources don't keep everyone else waiting. Make sure the new approach handles a mixture of slow and fast sources gracefully, maybe even allow fast sources to update while a slow source is still processing a previous iteration.