TST: regression test for concat with keys and mixed-freq PeriodIndex columns

[jbrockmendel]

Summary

• Closes BUG: Concat 2 dataframes with different PeriodIndex raises an exception when using keys parameter #51489.
• The originally-reported failure (pd.concat([q_df, y_df], keys=[...], axis=1) raising IncompatibleFrequency when the column PeriodIndex objects have different freq) no longer reproduces on main.
• It was fixed as a side effect of API: IncompatibleFrequency subclass TypeError #61875, which made IncompatibleFrequency subclass TypeError. The Categorical.__init__ factorize path that concat uses already had try: factorize(values, sort=True) except TypeError: factorize(values, sort=False), so the freq mismatch is now caught and falls back to an unsorted object-dtype level, which is the desired behavior.
• Adds a regression test in TestPeriodConcat covering the reproducer from the issue.

Test plan

• pytest pandas/tests/reshape/concat/test_datetimes.py::TestPeriodConcat -xvs

🤖 Generated with Claude Code

[mroeschke]

Thanks @jbrockmendel