[air] Minor Doc updates

[xwjiang2010]

Why are these changes needed?

Update a few docs and param names.

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 :(

[amogkam]

@xwjiang2010 I can use Predictor in piecemeal fashion by just instantiating it directly right?

create_compatible_checkpoint won't work for all Predictors since some Predictors expect a checkpoint dict and others expect checkpoint dir. I don't think we want to expose these details to the user and instead, only use Checkpoints that have been outputted by Trainers.

[amogkam]

Ah I see, the problem is not that I can't create the Predictor, but it's that I need a Checkpoint to actually use it with BatchPredictor or ModelWrapperDeployment.

In that case, I think we should explore this approach: #24977, which should be agnostic to the Checkpoint.

[xwjiang2010]

Ah got it, so the idea is that for non training case, user doesn't have to learn the Checkpoint concept, which makes sense. Let me update the PR to reflect that.
How about a Predictor.from_model_and_preprocessor sound?