Removed soft-deprecated code from posterior_vs_prior.

[tjmckinley]

Hi,

Thanks for developing rstanarm, it's made teaching Bayesian methods far easier and more accessible! I noticed a warning when running the posterior_vs_prior() function for more recent updates of tidyverse packages, because it relies on some soft-deprecated functions. A reprex is below:

## load libraries
library(rstanarm)

## set up data
dat <- data.frame(y = 8)

## fit binomial model
p_bayes <- stan_glm(
    cbind(y, 10 - y) ~ 1, 
    data = dat, 
    family = binomial(link = "logit"),
    prior_intercept = normal(0, 1.5)
  )

## compare 90% credible intervals for the prior and posterior
posterior_vs_prior(p_bayes, pars = "(Intercept)")

This returns:

Warning messages:
1: `aes_string()` was deprecated in ggplot2 3.0.0.
ℹ Please use tidy evaluation idioms with `aes()`.
ℹ See also `vignette("ggplot2-in-packages")` for more information.
ℹ The deprecated feature was likely used in the rstanarm package.
  Please report the issue at <https://github.com/stan-dev/rstanarm/issues>. 
2: The `size` argument of `grid_lines()` is deprecated as of bayesplot 1.16.0.
ℹ Please use the `linewidth` argument instead.
ℹ The deprecated feature was likely used in the rstanarm package.
  Please report the issue at <https://github.com/stan-dev/rstanarm/issues>. 

This PR contains a possible fix utilising rlang, which is a dependency of ggplot2, which itself is a dependency of rstanarm, so hopefully this should work without any further changes. Not sure if this is useful, but the code is here if it is. It seems to fix the problems for me.

Cheers,

TJ

[jgabry]

Thanks for developing rstanarm, it's made teaching Bayesian methods far easier and more accessible!

Thanks for the PR and for this comment. That was our goal when we developed the package!

Looking at your PR made me realize we have other deprecated ggplot2 syntax that we need to clean up so I'm going to push a few commits to this before merging.

@bgoodri while fixing deprecated ggplot2 syntax I also fixed a bug in plot_nonlinear so that it can now plot bivariate smooths from a model with more than one smooth term without an error.

[tjmckinley]

Awesome. Glad this was useful. Thanks again for the package.