C19 Update NN US and World

More progress on neural net, now generalized so I can do world, any country or any state. Turns out batch_size is an important parameter. These results use batch_size=1. However, it seems to my eye that the NN does not incorporate curvature of the last few data points in the projection. Experiments with various batch sizes show this is not the cause, although one would think a batch of say 5 would let the NN fit the last 5 points, pick up on the curvature, and use that to project forward. Not so. Curious.