motivate Cholesky factor parameterizations in functions reference and elsewhere

[bob-carpenter]

Summary:

We define the Cholesky-based parameterization, but don't mention that they're more efficient and more numerically stable.

Description:

This should go in the functions reference for the Cholesky-based parameterizations (LKJ, Wisharts, and multi-normal).

This arose through an issue on the Stan forums from jachymb, which also describes why the Cholesky parameterizations are better:

https://discourse.mc-stan.org/t/what-is-the-point-of-the-multi-normal-cholesky-parametrization/38483/2