Implement batched grid reduction #21
Description
Currently, grid_reduction = N builds up a dataset of size N one training point at a time. This maximizes the spacing of the points, but is slow when drawing N points from very large grids (> 1 million). Implement batch grid reduction as a keyword option grid_reduction_batch = 10. The spacing may be less robust, but the speed of selecting points will improve drastically.