Why are these changes needed?

The current implementation uses groupby(column).count() that causes a full sort. The new implementation uses AggregateFn which uses groupby(None) and set() to aggregate unique values.

The time complexity should be O(N / parallelism) according to ds.aggregate().

It's about 10x faster in my local test.

Some part of test_unique is removed because it was designed for the original implementation.

Related issue number

Closes #49298

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 :(