Backfill based on rows about polars HOT 1 OPEN

# This does not "scale"
backward_fill_horizontal = lambda cols: (
    pl.cum_reduce(function=lambda l, r: pl.coalesce(l, r), exprs=map(str, reversed(cols)))
      .struct.field("*")
)

import pandas as pd

df = pd.DataFrame({
    "instu_id": [9000098, 9000099],
    "COLL_END_DATE": ["2023-10-31","2023-10-31"],
    "Period": [1,3],
    "outstanding": [1.188779e+08, 1231231]
})

df["Period"] = pd.Categorical(temp_df["Period"], categories=[0, 1, 2, 3], ordered=True)

(df.groupby(["instu_id", "COLL_END_DATE", "Period"], dropna=False, observed=False)
   .agg(["any", "sum"])
) 

#                               outstanding
#                                       any          sum
# instu_id COLL_END_DATE Period
# 9000098  2023-10-31    0              NaN          0.0
#                        1             True  118877900.0
#                        2              NaN          0.0
#                        3              NaN          0.0
# 9000099  2023-10-31    0              NaN          0.0
#                        1              NaN          0.0
#                        2              NaN          0.0
#                        3             True    1231231.0