refactor(bigquery): move Arrow query result reading to _arrow_query

[ohmayr]

Moves the Arrow query results code out of RowIterator._download_arrow_from_job_id() into a new private module, _arrow_query.py, with no behavior change. The rest of the Arrow changes build on it, and _pyarrow_helpers.py is deprecated.

• RowIterator._download_arrow_from_job_id() keeps its signature and delegates to _arrow_query.download_arrow_from_job().
• Tests patch _arrow_query.pyarrow instead of table.pyarrow.

git diff --color-moved shows the moved block.

[gemini-code-assist]

Code Review

This pull request refactors the Arrow query results downloading logic by extracting the _download_arrow_from_job_id method from google/cloud/bigquery/table.py into a new dedicated module google/cloud/bigquery/_arrow_query.py. The unit tests have been updated accordingly to mock pyarrow in the new module. A review comment suggests validating that the project, location, and job ID are all present before constructing the stream name to prevent invalid resource paths and cryptic API errors.

[gemini-code-assist]

If row_iterator.client is None (which can happen if bqstorage_client is provided externally) or if the project, location, or job ID is missing, constructing the stream_name with None values will result in an invalid resource path and cause a cryptic API error. We should validate that project, location, and job_id are all present before constructing the stream name.

    project = row_iterator._project or (
        row_iterator.client.project if row_iterator.client else None
    )
    location = row_iterator._location or (
        row_iterator.client.location if row_iterator.client else None
    )
    job_id = row_iterator._job_id
    if not project or not location or not job_id:
        raise ValueError(
            "Project, location, and job ID must all be present to stream Arrow results."
        )
    stream_name = f"projects/{project}/locations/{location}/jobs/{job_id}/streams/_default"

[ohmayr]

I'd rather keep this PR a pure move with no behavior changes. On this path the project, location and job ID come from the job reference that query_and_wait() sets on the RowIterator, so they're populated in practice. If we want stricter validation it should be its own change.