Enable saving the random state, so that restarts are still reproducible

[bocklund]

ESPEI can be deterministic and reproducible, but restarting resets the random state.

That means running for 1000 steps in one run and two runs of 500 steps each (1000 total) will give different results, despite each being deterministic.

A solution is to be able to dump and load the random state on restart.

[bocklund]

It would also be nice if the probability and trace are bundled into whatever blob/format stores this information, so that restarts could be just pointing to the blob and the downstream user doesn't have to worry about concatenating probability/trace files.