Initializing joint marginals is pain

[bvdmitri]

We need an easy function to do that based on the following snippet I have for @ismailsenoz

function before_inference(model)
    graph = RxInfer.getmodel(model)
    gcv_labels = collect(filter(RxInfer.GraphPPL.as_node(GCV), graph))
    @info "Number of GCV nodes is $(length(gcv_labels))" 
    for label in gcv_labels
        factor_node = RxInfer.GraphPPL.getextra(graph[label], RxInfer.ReactiveMPExtraFactorNodeKey)
        for marginal in factor_node.localclusters.marginals
            _name = RxInfer.ReactiveMP.name(marginal)
            stream = RxInfer.ReactiveMP.getmarginal(marginal)
            if _name === :y_x
                RxInfer.ReactiveMP.setmarginal!(stream, MvNormalMeanCovariance([ 0.0, 0.0 ], [ 1.0 0.0; 0.0 1.0 ]))
            elseif _name === :z_κ
                RxInfer.ReactiveMP.setmarginal!(stream, MvNormalMeanCovariance([ 0.0, 0.0 ], [ 1.0 0.0; 0.0 1.0 ]))
            elseif _name === :ω
                RxInfer.ReactiveMP.setmarginal!(stream, NormalMeanVariance(0.0, 1.0))
            end
        end
    end
end

Also GraphPPL exposes NodeId plugin, which is not used in RxInfer but might be very handy here