Revert "Revert "[Data] Implement Operators for union()""

[scottjlee]

Reverts #36583, originally implemented in #36242. The changes made in the original PR seem unrelated to the failures on linux://python/ray/dashboard:modules/job/tests/test_http_job_server (:serverless: Dashboard Tests: test_http_job_server.py:test_ray_tune_basic). In the revert PR CI tests, we see the same issue, so I'm thinking that the original PR is not the culprit for the underlying test failure.

E       Traceback (most recent call last):
--
  | E         File "ray_tune_basic.py", line 14, in <module>
  | E           tune.run(objective)
  | E         File "/ray/python/ray/tune/tune.py", line 965, in run
  | E           callbacks = _create_default_callbacks(
  | E         File "/ray/python/ray/tune/utils/callback.py", line 141, in _create_default_callbacks
  | E           callbacks.append(TBXLoggerCallback())
  | E         File "/ray/python/ray/tune/logger/tensorboardx.py", line 173, in __init__
  | E           from tensorboardX import SummaryWriter
  | E         File "/opt/miniconda/lib/python3.8/site-packages/tensorboardX/__init__.py", line 5, in <module>
  | E           from .torchvis import TorchVis
  | E         File "/opt/miniconda/lib/python3.8/site-packages/tensorboardX/torchvis.py", line 11, in <module>
  | E           from .writer import SummaryWriter
  | E         File "/opt/miniconda/lib/python3.8/site-packages/tensorboardX/writer.py", line 17, in <module>
  | E           from .comet_utils import CometLogger
  | E         File "/opt/miniconda/lib/python3.8/site-packages/tensorboardX/comet_utils.py", line 6, in <module>
  | E           from .summary import _clean_tag
  | E         File "/opt/miniconda/lib/python3.8/site-packages/tensorboardX/summary.py", line 13, in <module>
  | E           from .proto.summary_pb2 import Summary
  | E         File "/opt/miniconda/lib/python3.8/site-packages/tensorboardX/proto/summary_pb2.py", line 16, in <module>
  | E           from tensorboardX.proto import tensor_pb2 as tensorboardX_dot_proto_dot_tensor__pb2
  | E         File "/opt/miniconda/lib/python3.8/site-packages/tensorboardX/proto/tensor_pb2.py", line 16, in <module>
  | E           from tensorboardX.proto import resource_handle_pb2 as tensorboardX_dot_proto_dot_resource__handle__pb2
  | E         File "/opt/miniconda/lib/python3.8/site-packages/tensorboardX/proto/resource_handle_pb2.py", line 36, in <module>
  | E           _descriptor.FieldDescriptor(
  | E         File "/opt/miniconda/lib/python3.8/site-packages/google/protobuf/descriptor.py", line 561, in __new__
  | E           _message.Message._CheckCalledFromGeneratedFile()
  | E       TypeError: Descriptors cannot not be created directly.
  | E       If this call came from a _pb2.py file, your generated code is out of date and must be regenerated with protoc >= 3.19.0.
  | E       If you cannot immediately regenerate your protos, some other possible workarounds are:
  | E        1. Downgrade the protobuf package to 3.20.x or lower.
  | E        2. Set PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python (but this will use pure-Python parsing and will be much slower).
  | E
  | E       More information: https://developers.google.com/protocol-buffers/docs/news/2022-05-06#python-updates

[c21]

LGTM, assuming no new code change compared to original PR.