[docs] Remove trainer references from preprocessors

[richardliaw]

Why are these changes needed?

This PR contains the following changes:

1. Removes a lot of the AIR trainer <> preprocessor usage and lifecycle text that is no longer relevant
2. Adds references and recommendation to the map_batches approach
3. Also adds hooks from Ray Train docs (especially tabular-oriented trainers like XGBoost).

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 :(

[amogkam]

Can we update the wording here? It's a little bit confusing right now if we say the recommended way for preprocessing is existing Ray Data operations, but then say you should consider built-in preprocessors.

Maybe just "While Ray Data supports generic transformations on datasets, for tabular data it also provides built-in preprocessors"

[amogkam]

Suggested change

	Preprocessing Data
	Preprocessing Tabular Data

[pcmoritz]

We should also update the example https://docs.ray.io/en/master/ray-air/examples/gptj_deepspeed_fine_tuning.html to remove the preprocessor stuff and instead use map_batches :)