[RLlib] Fix gym.Wrapper problem for MA Envs.

[sven1977]

When wrapping a MultiAgentEnv with a gym Wrapper, RLlib incorrectly pre-processes observations and crashes due to a Space mismatch. This PR makes sure that RLlib is always aware of the actual underlying (wrapped) Env.

Fixes issue #8303

Closes #8303

• I've run scripts/format.sh to lint the changes in this PR.
• I've included any doc changes needed for https://docs.ray.io/en/latest/.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failure rates at https://ray-travis-tracker.herokuapp.com/.
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested (please justify below)

[AmplabJenkins]

Can one of the admins verify this patch?

[ericl]

You can't use a reward wrapper around a MultiAgentEnv though right? I don't think this is a valid use case.

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/25536/
Test PASSed.

[sven1977]

@ericl It would be intuitive, though. Can we merge this either way? We don't have to officially support it, but it would help a lot of users who are simply doing this for convenience and expect this to work fine (it does, after all).

[ericl]

I'm concerned this is a hacky behavior though. It doesn't really make sense to crawl the inheritance chain looking for a subclass signature. Also, there are clear workarounds to use non gym wrappers.

I think it would be asking trouble to start half supporting gym wrappers for multiagent envs, which were never intended to be compatible anyways.