[Data] Account internal inqueue to previous operator's memory usage

[bveeramani]

Why are these changes needed?

To make informed scheduling decisions, Ray Data estimates the object store memory usage from each operator. Currently, an operator's inputs count towards its memory usage; however, this can lead to suboptimal scheduling decisions since we primarily care about outputs for backpressuring.

This PR updates the implementation so that input counts toward the previous operator's object store usage (i.e., in counts towards the operator that created the data).

Related issue number

Required for #43026

See also #42851 (comment)

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 :(