Bug Report - SchemaError with apply Function in Polars 0.20.3

[maazizzahra]

Checks

• I have checked that this issue has not already been reported.
• [] I have confirmed this bug exists on the latest version of Polars.

Reproducible example

lf = pl.LazyFrame({'a': ['[]']})
lf.select(pl.col('a').apply(eval, return_dtype=pl.List(pl.Utf8))).collect()
Polars + moderne :
lf.select(pl.col('a').map_elements(eval, return_dtype=pl.List(pl.String))).collect()
-> crash

I encountered the following error: polars.exceptions.SchemaError: expected output type 'List(String)', got 'List(Null)'; set return_dtype to the proper datatype
To resolve the issue, I used the following workaround:

lf.select(pl.when(pl.col('a') == '[]').then(pl.lit([]).cast(pl.List(pl.String))).otherwise(pl.col('a').map_elements(lambda x: eval(x) or [''], return_dtype=pl.List(pl.String)))).collect()

we use  Polars 0.20.3 

Installed versions

I use 0.20.3```not 0.20.1

</details>

[MarcoGorelli]

thanks for the report, could you show an example of input data please

[cmdlineluser]

Is this the same issue that is discussed in #10398 (comment)?

i.e. List(null) in this case should satisfy/upcast to List(String)

You can also force the dtype by manually wrapping in a Series inside the callback.

lf.select(pl.col('a').map_elements(eval)).collect()
# shape: (1, 1)
# ┌────────────┐
# │ a          │
# │ ---        │
# │ list[null] │ # <- null
# ╞════════════╡
# │ []         │
# └────────────┘

lf.select(
    pl.col('a').map_elements(lambda x: pl.Series(eval(x), dtype=pl.String))
).collect()
# shape: (1, 1)
# ┌───────────┐
# │ a         │
# │ ---       │
# │ list[str] │ # <- str
# ╞═══════════╡
# │ []        │
# └───────────┘

[stinodego]

Should be fixed by #15699