Divergences for mcmc_scatter

[jgabry]

This PR adds an np and np_style arguments to mcmc_scatter so that divergences can be highlighted in the plot. It also adds the scatter_style_np function to use when specifying np_style.

(It's a little bit annoying to have a bunch of functions (e.g. scatter_style_np, pairs_style_np, trace_style_np, parcoord_style_np) that are used to specify the np_style argument depending on the plotting function. It would be nice just to have one function for that, but I tried playing around with that and it's even more annoying because it makes it more confusing to have different defaults for the shape, style, alpha, arguments depending on the plotting function. With different functions the default values can be made more explicit.)

Example usage

library(rstan)
fit <- stan_demo("eight_schools")
posterior <- as.matrix(fit)

color_scheme_set("darkgray")
mcmc_scatter(
  posterior, 
  pars = c("tau", "theta[1]"), 
  np = nuts_params(fit), 
  np_style = scatter_style_np(div_color = "green", div_alpha = 0.8)
)

[jgabry]

(Running R CMD CHECK now, will report back with the results)

[jgabry]

Passes R CMD CHECK

[tjmahr]

This looks good. We'll probably need a vignette to show off all these _np functions.

[jgabry]

Cool. I just added a quick example of using mcmc_scatter with the np argument in the visual-mcmc-diagnostics.Rmd vignette. In the previous PR for mcmc_parcoord I also added an example of that to the same vignette. Eventually it will probably make sense to break up the current vignettes into smaller ones if/when they get too long.