[RLlib] SampleBatch: Docstring- and API cleanups; Add support for nested data.

[sven1977]

SampleBatch: Docstring- and API cleanups; Add support for nested data.

• adds support for SampleBatches to hold nested data, e.g. a dict of dicts, of tuples of np.ndarrays. All methods have been enhanced to use the dm_tree module when having to loop through the batch. This will allow us later to handle individual "obs" (or other) sub-components in case users want to specify view requirements for only parts of the observation (e.g. obs_space=Dict(img=..., level=...) -> framestack only the img component of the observations via trajectory view API).
• clean up SampleBatch API further (deprecation warnings, rename zero_pad into right_zero_pad for clarity, etc..).
• add more SampleBatch test cases.
• add slicing operator support: e.g. my_slice = batch[4:102].

Why are these changes needed?

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 :(