This does not actually work, but I'm leaving it around because maybe it'll work in Java 8.
This way unitialized fields do not show up in XML dumps.
Unless fixedStepDt was set properly, ExactStochastic would needlessly stop every fixedStepDt. It needs only to be small enough to enable proper reporting, can can be the whole runtime if there is no reporting. SteppepedStochastic does not really work if fixedStepDt is not an even divisor of the reporting time, and will return bogus results for the last value. Not easy to fix because log(dt) is stored in various places.
becausing calling it calculation seed in places was just confusing.
It seems that a) using large chunksize is not beneficial b) interleaving getData() with other operations leads to corruption and slowdown c) the first row in each extensible array had bogus results since always With current changes, the slowdown from one writer is on the order or single percent, and with 8 writers it is on the order or 10-20%, which should be tolerable for now.
The time and speed ignore the initial setup time, so they are meaningful for short runtimes even with big models where the setup if fairly expensive.
Previous version was just buggy, and was failing with Asia's model.
Basically, the array was written one event to late.
When running multiple trials in parallel, those calculations were needlessly repeated, and because those calculations actually update the state of the source objects, they weren't idempotent. Fix both problems by doing them before any parallel calculations start.