Splits the tuple-names → MultiIndex columns fix out of GH-65862 into a standalone, bite-size change.

With engine="pyarrow", passing tuples in names produced flat columns (an Index of tuples) instead of MultiIndex columns like the C and python engines:

>>> pd.read_csv(StringIO("1,2\n3,4"), names=[("a", "x"), ("b", "y")], engine="pyarrow").columns
Index([('a', 'x'), ('b', 'y')], dtype='object')   # was flat; other engines give a MultiIndex

The fix routes the result columns through the same _maybe_make_multi_index_columns helper the C/python engines already use, so all engines now agree.

• 4-line fix in arrow_parser_wrapper.py (no-op when columns aren't tuples)
• un-xfails the names-tuple parametrizations of test_header_multi_index_common_format1/2/3 — their blanket @xfail_pyarrow only applied because the separate header=[0, 1] parametrization raises; those now assert the correct MultiIndex, with the list-header variant kept as a conditional xfail (that's a separate concern handled in BUG: raise helpful errors for unsupported pyarrow read_csv usage #65862)
• adds a dedicated regression test

🤖 Generated with Claude Code