[Data] Update Data progress bars to use row as the iteration unit

[scottjlee]

Description

As a followup to #46432, we want to further improve clarity of the progress bar output. Currently, we use bundle (corresponding to RefBundle) and block (corresponding to Block) as the iteration unit, which are both internal concepts that some Ray Data users may be unfamiliar with.

With some more involved code changes, we can replace these with row, corresponding to rows in the output Dataset. This is the most atomic unit of the Dataset that all users should be well familiar with, since rows are a fundamental concept in almost all data processing libraries.

Code pointers:

• MapOperator._task_done_callback(): Add logic to estimate number of output rows from operator, based on number of rows from completed tasks so far. Can follow existing logic to estimate number of output bundles.
• Add a method/property PhysicalOperator.estimated_output_num_rows to get the number of estimated rows described above.
• Use the op.estimated_output_num_rows to update progress bars:
  • https://github.com/ray-project/ray/blob/master/python/ray/data/_internal/execution/streaming_executor_state.py#L211
  • https://github.com/ray-project/ray/blob/master/python/ray/data/_internal/execution/streaming_executor.py#L149

Use case

Further improve clarity of Ray Data progress bar