Poor support of using column in IO

[xukai92]

Model using : below is poorly supported to fetch variables from Chain.

@model ldamodel(K, V, M, N, w, doc, alpha, β) = begin
  theta = Matrix{Real}(M, K)
  for m = 1:M
    theta[m,:] ~ Dirichlet(alpha)
  end

  phi = Matrix{Real}(K, V)
  for k = 1:K
    phi[k,:] ~ Dirichlet(β)
  end

  # z = tzeros(Int, N)
  # for n = 1:N
  #   z[n] ~ Categorical(theta[doc[n]])
  # end

  for n = 1:N
    w[n] ~ Categorical(phi' * theta[doc[n],:])
  end

end

Have to write variables in type of nest arrays at this moment.

[yebai]

I thought this before, but didn't manage to find a good solution. Another possible use case:

x = Array(8,8)
x[:, 1:4] ~ Normal

[xukai92]

Note: using matrix is the vectorisation way actually works around this problem, BUT users need to fetch variables using the nested-array syntax:

@model xxx() = begin
  m = Matrix(2,2)
  m ~ [MvNormal(...)]
end

chn = sample(...)
chn[Symbol("m[1][1]")] # this works

[yebai]

Fixed by #295