Fix baseline optimization order

[ryanjulian]

The baseline should be optimized before the policies, because the augmented reward and return values depend on network parameters