[WIP] Don't need to calculate cdf with np.where to plot p-p plot

[vhaasteren]

Don't merge this yet, I first want to have the discussion.

I've been wondering about the following. When making a p-p plot, la forge and some other packages I've seen calculate the cdf by counting realizations with up to a certain p-value. But the p-value is supposed to be distributed according to Uniform(0, 1). That means you can just straight up plot p-value vs q. Compared to the 'counting' way, this means you have slightly more independent datapoins in the p-p plot.

See what I mean at the code diff below. The difference is very minimal, but it's more elegant.

Is there a reason not to be changing this?