[RLlib] Fix TorchMultiCategorical.to_deterministic to return a multi-dimensional tensor instead of a list.

[simonsays1980]

Why are these changes needed?

Using a MultiDiscrete action space results in RLlib in a MultiCategorical distribution. When turning this distribution to a deterministic distribution for inference, a list of TorchDeterministic distributions is created. This results than in a sampled list of tensors instead of a single tensor with proper dimensions. This PR proposes a fix that

• Turns a TorchMultiCategorical distribution into a single TorchDeterministic that returns a single tensor of proper distribution.
• It works also for MultiDiscrete distributions with different categories per dimension, by using padding.

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 :(

[sven1977]

LGTM now. Thanks for this fix @simonsays1980 !