Similarly as for normal reductions (eg #30982), we should investigate having masked array-specific support in the groupby algorithms.

Currently, when starting from a nullable extension array, they get converted to a numpy array (eg integers with missing values will typically get cast to float with nan) before passing to the cython algorithm.
Having support for passing a mask to the cython algos can improve the groupby support for nullable dtypes.

• any / all (correctly pass mask + add Kleene logic) -> ENH/BUG: Use Kleene logic for groupby any/all #40819
• cummin / cummax -> PERF/BUG: use masked algo in groupby cummin and cummax #40651
• cumsum / cumprod -> ENH: Support mask in GroupBy.cumsum #48070, ENH: Support mask in groupby cumprod #48138
• sum / prod / var / mean
• min / max -> BUG: Groupby min/max with nullable dtypes #42567
• median
• ohlc (ENH: Add support for groupby.ohlc for ea dtypes #48081)
• quantile (correctly pass mask)
• last / nth (last PERF: support mask in group_last #46107)
• rank (ENH: support mask in libalgos.rank #46932)