feat: Add ignore_nulls for pl.concat_str

[reswqa]

I set the default value of ignore_nulls to False so that we can keep the same behavior as before. We could change it to True in the breaking release then.

[reswqa]

By offline discuss with Alex. This SQL part change the behavior of Concat and ConcatWS, but they didn't match the SQL standard before indeed. So this changes should be treated as a bug-fix.

[reswqa]

For concat + str and str + concat, we can only simplify it if concat propagate null values, otherwise it will change the behavior of None + str = None.

For example, the following result should be a Series with None:

pl.select(
      pl.lit(None, dtype=pl.String) 
     + pl.concat_str(pl.lit("a"), pl.lit("b"), ignore_nulls=True)
)

[reswqa]

@alexander-beedie, I think that's the kind of behavior we want, right?

[mcrumiller]

Is there an issue with a discusson related to this? I never realized that having a single null value wipes your entire concat_str string:

>>> pl.select(pl.concat_str(pl.lit("a"), None, pl.lit("b")))
shape: (1, 1)
┌─────────┐
│ literal │
│ ---     │
│ str     │
╞═════════╡
│ null    │
└─────────┘

I would prefer a null_str='' argument, which allows for cases like:

>>> pl.select(pl.concat_str(col("last_name"), col("first_name"), null_str="<missing>", separator=", "))
shape: (1, 1)
┌──────────────────┐
│ literal          │
│ ---              │
│ str              │
╞══════════════════╡
│ Smith, John      │
| Price, <missing> |
└──────────────────┘

I suppose in the meantime one could do col(x).fill_null("<missing>"). Maybe this has to do with aligning with SQL behavior, which I'm unsure how exactly that works.

[reswqa]

I never realized that having a single null value wipes your entire concat_str string.

Unfortunately, this is the default behavior right now, but I want to change it in the next breaking release.

Is there an issue with a discusson related to this?

Yes, we have #13633.

I would prefer a null_str='' argument, which allows for cases like:

I tend to explicit use col(x).fill_null(), not by an extra keyword argument. ignore/propagate nulls are more basic behaviors that the engine should do.

[mcrumiller]

@reswqa makes sense, thanks!