Align RL and IL evaluation environment code

[StoneT2000]

currently some do partial resets, some don't, and some report different version of success/success_once/success_at_end. This needs to be aligned to be just one setting:

• no partial resets (evaluation until max episode steps)
• report success_once and success_at_end instead of an ambiguous success metric
• update docs on evaluation in the RL and IL guides