[Data] Try get size_bytes from metadata and consolidate metadata methods

[bveeramani]

Why are these changes needed?

LogicalOperator and Datasource expose methods like schema() to make metadata efficiently available to Dataset APIs like Dataset.schema(). Currently, LogicalOperator exposes three such methds: num_rows(), schema(), and input_files().

This PR adds size_bytes() because it was missing. To simplify the interface, it also consolidates the metadata methods into a single aggregate_output_metadata() method.

Related issue number

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 :(

[scottjlee]

nit: not a big deal especially since it elongates the code. technically this is a double for loop, which could be optimized into one loop and early exit if we run into a Non num_rows. something like:

sum_num_rows = 0
for bundle in self.input_data:
    if not bundle.num_rows():
        return None
    sum_num_rows += bundle.num_rows()
return sum_num_rows

and the same could be applied in other similar places, e.g. _size_bytes method and methods in AbstractFrom. it may be worthwhile to put this logic in a shared utility function, for example as a static method in BlockMetadata.

[bveeramani]

Makes sense, although I prefer keeping as-is since we don't know if this is a performance issue and I want to avoid premature optimization