I have a dataset with N number of columns, let’s say 3 to generate an example (n_1, n_2, n_3).
There are duplicate values based on some columns, as in:

import polars as pl

data = {
    'n_1': ['a', 'b', 'a', 'c', 'd', 'b', None, 'c', 'e'],
    'n_2': [1, 2, None, 3, 1, 2, 1, 3, 5],
    'n_3': [123, 345, 123, 567, 123, 987, 123, None, 923]
}
df = pl.DataFrame(data)

┌──────┬──────┬──────┐
│ n_1  ┆ n_2  ┆ n_3  │
│ ---  ┆ ---  ┆ ---  │
│ str  ┆ i64  ┆ i64  │
╞══════╪══════╪══════╡
│ a    ┆ 1    ┆ 123  │
│ b    ┆ 2    ┆ 345  │
│ a    ┆ null ┆ 123  │
│ c    ┆ 3    ┆ 567  │
│ d    ┆ 1    ┆ 123  │
│ b    ┆ 2    ┆ 987  │
│ null ┆ 1    ┆ 123  │
│ c    ┆ 3    ┆ null │
│ e    ┆ 5    ┆ 923  │

I want to fill the null values in each column with the value that is present in similar records, so that e.g. row 3 and 7 become ['a', 1, 123], row 8 becomes ['c', 3, 567].

EDIT: I forgot to add that that row 7 could become two rows and fill the values with both 'a' and 'b' so you have both [d, 1, 123] and [a, 1, 123].

Do you know the optimal strategy to achieve this in plain SQL or Polars or Spark?
Thanks