[RLlib] Use SampleBrach instead of input dict whenever possible

[gjoliver]

Why are these changes needed?

To get rid of all the SampleBatch["is_training"] deprecated warnings.

Related issue number

Checks

• 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 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]

Nice!

[sven1977]

Great fix @gjoliver . Just one nit before we merge. I know the SampleBatch c'tor is flexible, but we should probably use only either "_is_training" or "is_training" in all (new) SampleBatch instantiations.

[gjoliver]

Great fix @gjoliver . Just one nit before we merge. I know the SampleBatch c'tor is flexible, but we should probably use only either "_is_training" or "is_training" in all (new) SampleBatch instantiations.

totally. I debated a little about whether I should make the ctor parameter is_training. so went back and forth. probably missed a few places when I renamed them.
thanks for catching them.