state_normalization = variance does not work with fd rnn trainning

[Neo-X]

The fd model will have the default [[0, ..., 0], [1, ...., 1]] state bounds.

[Neo-X]

The bootstrapping process only returns state bounds for the policy. If the fd uses a different kind of state representation it will not work. Currently setting the fd bounds this way is not done at all. Use the adaptive method for now.