Deploy a trained recurrent policy

[dtbpkmte]

How can I deploy a trained recurrent policy, in a way similar to #87? The instructions in #87 works perfectly for non-recurrent policies, but the recurrent policies also need the hidden state updates. I can add these logics to my Actor class, but I am wondering if there is a simpler way that I overlooked. Thank you!