RNG should support concurrent generation

[kyonifer]

For some applications such as Monte-Carlo, random number generation can be the bottleneck in performance. It would be useful to support a concurrent RNG that can generate numbers in multiple threads (probably using a fork join pool and/or coroutines) using all available cores and zip them back into a single stream for the requesting thread. This would probably only make sense for large arrays, so a hybrid single/multi-threaded approach may be appropriate.