Performance Optimizations

[aslanides]

Here are some easy algorithmic improvements as well as performance tuning/optimization.

1. I'm pretty sure we're re-drawing all of the plots at each iteration, which is unnecessarily quadratic. I think there's a big-ish speed-up to be had here (~30%) given how slow the real-time UI rendering currently is.
2. Dirichlet model updates are O(1) (great!), but save() and load() are still O(N^2). I think we can get these down to O(m) (MCTS horizon) pretty easily, and that'll be a big win for environment scaling.

Check for V8 optimizations throughout using the profiler. Apparently it's not happy with let keywords, and can't optimize code that uses them.

[aslanides]

Dirichlet model is greatly improved now, with significantly faster run-time than the standard Bayes mixture now. All the gridsize-dependent scaling has been removed, and so now we can run this model on bigger and more interesting grids.

As for the plots: I've looked into this is a little bit, and it seems like a chunk of work to go through the d3 documentation to figure out how to re-write this. I'm not even sure if this is a bottleneck now -- it's hard to profile. Todo :)