Make Turing CI pass with VNT

[penelopeysm]

This is underway in TuringLang/Turing.jl#2756, just opening an issue so that I can track it, since we can't release 0.40 until we're confident that Turing is fine

[penelopeysm]

Update: After we got transform strategies encoded properly, now MH finally works correctly!

using Turing
@model function f()
    x ~ Beta(2, 2)
    y ~ Beta(2, 2)
end
model = f()
chain = sample(model, Gibbs(:x => MH(), :x => MH(:x => LinkedRW(0.5)), :y => MH()), 10000)

julia> mean(Beta(2, 2)), std(Beta(2, 2)) # Expected values
(0.5, 0.22360679774997896)

julia> mean(chain[:x]), mean(chain[:y])
(0.49931294879210025, 0.5011466846935558)

julia> std(chain[:x]), std(chain[:y])
(0.22204883745457468, 0.22357913987510789)

[penelopeysm]

OK, we're almost there. It's just GibbsConditional that needs updating now.

[penelopeysm]

Done