Better Handling for MCMC Proposal Distributions

[jxwuyi]

Currently we did a hack for gaussian proposals.

The proposal distribution can be found in "MCMC_util.h" called "gaussian_prop". Note that the variance of the proposal distribution is fixed: 1.0.

The general framework in Swift goes as follows:
In GibbsTranslator (only did for Gibbs now), when conjugacy analysis fails and the return type is Real, the compiler will automatically use the gaussian_proposal.
Basically it calls the general sample method "mh_symmetric_resample_arg(node, g)" in "MCMC.h". This method firstly takes in a BayesVar, then the second argument is a symmetric proposal distribution g, i.e., g(x->y) = g(y->x).

It now has the following issues:
(1) what if the real variables have bounded support??????
(2) what about multi-gaussian?
(3) how to allow the user to specify whether (not) to use gaussian proposal? how to specify the variance???

Ideally:
we should allow user to specify any customized proposal for every random variable via an external config file.

This issue remains open as a reminder for this hack. This hack is primarily for PPAML Challenge Problems.