Missing rand procedure for MvNormalMeanScalePrecision

[albertpod]

Following the discussion, we have found that sampling from MvNormalMeanScalePrecision is not implemented, i.e.

julia> dist = MvNormalMeanScalePrecision(zeros(2), 2.0)
MvNormalMeanScalePrecision{Float64, Vector{Float64}}(μ=[0.0, 0.0], γ=2.0)

julia> rand(dist)
ERROR: MethodError: no method matching length(::MvNormalMeanScalePrecision{Float64, Vector{Float64}})

[bvdmitri]

Maybe we should move this distribution in ExponentialFamily.jl?

[wouterwln]

Fixed by #206