[tune] populate internal configs when creating Trainable through DistributedTrainableCreator

[matthewdeng]

Why are these changes needed?

DistributedTrainableCreator creates a wrapper Trainable which is used to create an ImplicitFunc. The ImplicitFunc should be created with the same configs as the Trainable, but is currently not because Trainable.__init__ clears out some of the internal configs (e.g. trial info).

Changes

1. Create a new DistributedTrainable class with a build_config method to populate internal configs.
2. Update {horovod, tensorflow, torch} integrations to inherit from DistributedTrainable and call build_config.

Related issue number

Closes #15288

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]

Thanks a bunch for cleaning this up and adding tests! Looks good overall. Just left 1 comment.

[richardliaw]

Nice work! Left 1 nit.

[amogkam]

Nice, LGTM!