More operations for `GraphInfo`

[yebai]

In #47, we implemented a few basic operations for GraphInfo, for example, construction, getindex. For practical purposes, we need to be able to

• sample from GraphInfo model, i.e. sample all stochastic nodes from their prior distribution
• condition/decondition, i.e. fix values of certain nodes to user-specified value
• logdensityof, i.e. compute the log joint probability of a GraphInfo model given values of all nodes.

A good testing case for these features would be

1. define a model
2. initialise the model by sampling from the prior
3. compute the log density of the sampled parameters

[yebai]

@PavanChaggar this would be a nice follow-up PR for https://github.com/TuringLang/AbstractPPL.jl/pull/47, but supporting sampling depends on #51 I think.

[phipsgabler]

Transferred to SymbolicPPL.jl