Scaling `sigma` to match actuator `forcerange`/`ctrlrange`

[bbokser-ai]

I had issues getting my simulated robot to actuate with more than ~1-5 Nm on actuators that had torque limits (described in the MJCF via the forcerange parameter) on the order of ~70 Nm. Turned out that tuning the sigma_max argument when calling CrossEntropyMethodConfig() fixed the issue. I had assumed Judo would automatically scale its control signal ranges to match actuator limits, but it looks like this is not the case.

It might be good to have such an option, based on actuator limits from the MJCF, or at least make it clearer in the documentation that this needs to be tuned.

[bbokser-ai]

Update on this: it looks like setting action_normalizer="min_max" in ControllerConfig() does this by looking at model.actuator_ctrlrange. This works better than tuning sigma because it accounts for individual actuator limits. Leaving this issue up for the documentation aspect.