[Feature Request] Remove `eval_env` parameter

[araffin]

🚀 Feature

eval_env and eval_freq parameters were new in SB3 and were meant to facilitate evaluation in a separate env, by automatically creating evaluation env and EvalCallback. However, they don't seem to be used that much and don't offer the flexibility of the EvalCallback at the cost of code duplication and maybe confusing parameters.

The idea would be to first deprecate (next SB3 version) and then remove all the parameters related to that feature. Main drawback is that this would be a breaking change for people implementing custom algorithms.

### Checklist

• [x] I have checked that there is no similar issue in the repo (required)

[tobirohrer]

Hey, I wanted to check if this change is still planned. If so, I would take a look at it in the coming days.

[araffin]

Hello, it is still planned, but it's not a priority. I would like to integrate LN and dropout (https://github.com/DLR-RM/stable-baselines3/issues/1069), and probably progress bar callback first (https://github.com/DLR-RM/rl-baselines3-zoo/pull/287). Actually, I would welcome more help to improve the RL Zoo.

However, you can of course work on such PR, please only deprecate the feature first (it should output a warning and remove the feature).