add Graph variable to support Discrete variables sampled from their neighbors

[kunyuan]

Some discrete variable is better sampled from a set of neighbors instead of sampling globally. Once such discrete variable is implemented, there is no need for the MC update changeOrder in :mcmc. One can merge :mcmc solver with :vegasmc.

Once this is implemented, it probably makes sense to rename :vegas to :mc and :vegasmc to :mcmc. Indeed, Vegas is the name for the adaptive map used in the continuous variable. It is better not to use it as a name for the entire MC algorithm.

[kunyuan]

It is better to introduce a new Graph variable. The Markov-chain to perform a random walk on the Graph to sample a discrete variable