pandas 3 removed the M/H/T/Y/Q/BM frequency aliases; Period still wants 'M'

pandas 2.2.3 accepts resample('M'), 'H', 'T', 'Y', 'Q', 'BM' with only a FutureWarning. pandas 3.1.0 raises ValueError 'Invalid frequency: M' in resample and date_range. Replace them with 'ME', 'h', 'min', 'YE', 'QE', 'BME'; results are identical on 2.2.3, so the change is safe to make before upgrading. The trap: Period frequencies did not change. pd.period_range(..., freq='M') still works, and freq='ME' raises ValueError for periods, so a blind search-and-replace of 'M' to 'ME' breaks PeriodIndex code. Timedelta units are on a different schedule: pd.Timedelta(1, 'H') still works on 3.1 with a Pandas4Warning. Unchanged in both versions and still a silent trap: a groupby sum over an all-NaN group gives 0.0 (use min_count=1 for NaN), and NaN group keys are dropped unless dropna=False.