fix: step type in autoreset wrapper

[dluo96]

Is your feature request related to a problem? Please describe

When using AutoResetWrapper, the sequence of timestep.step_type returned during a rollout shows 0 values where the env has been auto reset instead of 2 values. Recall that

• 0 corresponds to the first step
• 1 corresponds to a mid step
• 2 corresponds to the last step

Describe the solution you'd like

Move the auto reset bug: [1, 1, 1, 2, 0, 1, 1, 1] has to be converted to [1, 1, 1, X, 1, 1, 1] with X being 0 or 2. Right now it is 0 but it should be 2 to warn the user that the episode got terminated. This would be important e.g. if using while not timestep.last().

timestep = timestep.replace(  # type: ignore
    observation=reset_timestep.observation
)

[clement-bonnet]

Got fixed in #123