Things to fix in ADVI

Summary:

These were issues that were noticed when code reviewing for #1751. @bob-carpenter spotted most of these.

Description:

• Change gradient_samples argument to the algorithm to something closer to "size of the sample" or "number of draws"
• in fullrank and meanfield, rather than cont_params.resize(cont_vector.size());, we should use an Eigen::Map
• document @throw and @return properly. Currently, run() looks like it can only return 0, but it does throw.

Current Version:

v2.14.0

Start by locating the ADVI algorithm entry point and the implementations of fullrank and meanfield, then inspect gradient_samples and the cont_params.resize(cont_vector.size()) call. Review run()'s documented return and exception behavior. Done means the argument naming, Eigen mapping, and @throw/@return documentation items are all addressed and the relevant checks pass.