[RLlib] Add experimental option to suppress algo config validate(); some code cleanups.

[sven1977]

• Add experimental option to suppress config validation. This will allow us to make users pay attention to raised config validation errors, but at the same time give them an opt-out in case they want to use a (bad) setting no matter what. Example: Running APPO with GPU and 1 remote learner is bad. Or running APPO with CPU and 1 local learner is very slow (but might be useful for debugging).
• Some other code cleanups.

Why are these changes needed?

Related issue number

Checks

• I've signed off every commit(by using the -s flag, i.e., git commit -s) in this PR.
• 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/master/.
  • I've added any new APIs to the API Reference. For example, if I added a
    method in Tune, I've added it in doc/source/tune/api/ under the
    corresponding .rst file.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested :(

[simonsays1980]

LGTM. A small question to "auto" rollout_fragment_length.

[simonsays1980]

Do we only support "auto" now?

[sven1977]

No, just the check for train_batch_size < 0 - now that we don't have the DDPPO algo anymore - seemed superfluous. So I took out that one check.