fix: output name of ternary expr

[nameexhaustion]

Fixes issue #12380 (introduced by #12357).

The previous PR lost the series name and dtype for the non-matching branch, but they need to be preserved to have the correct output name and dtype even if the values are never used.

[ritchie46]

Can we write this in terms of if else.

[nameexhaustion]

yeah

[ritchie46]

idem

[nameexhaustion]

@ritchie46 - just had a realization that it is probably not a good idea to be broadcasting to the matching branch... I think instead it's better to maintain the previous behavior (i.e. require all branches to have the same height and error otherwise) - otherwise we get:

df = pl.DataFrame(
    {"x": range(5), "y": range(5, 10)}, schema={"x": pl.Int8, "y": pl.Int64}
).with_columns(true=True, false=False)

df.select(pl.when(pl.first("true")).then(pl.col("x")))
# shape: (5, 1)
# ┌─────┐
# │ x   │
# │ --- │
# │ i8  │
# ╞═════╡
# │ 0   │
# │ 1   │
# │ 2   │
# │ 3   │
# │ 4   │
# └─────┘
df.select(pl.when(pl.first("false")).then(pl.col("x")))
# we get this if we broadcast to the matching branch (on 0.19.13):
# >>> df.select(pl.when(pl.first("false")).then(pl.col("x")))
# shape: (1, 1)
# ┌──────┐
# │ x    │
# │ ---  │
# │ i8   │
# ╞══════╡
# │ null │
# └──────┘

# but it should probably instead be this (on 0.19.12):
# shape: (5, 1)
# ┌──────┐
# │ x    │
# │ ---  │
# │ i8   │
# ╞══════╡
# │ null │
# │ null │
# │ null │
# │ null │
# │ null │
# └──────┘

So I think instead it's probably better to have #12357 to be reverted instead of merging this (or I can adjust this PR as well to remove the special broadcasting) - @JulianCologne raises a good point in #12363, a better fix for the original issue (#12354) is probably to have .first / .last always return a 1-item Series instead.

*as a note, the current change in behavior in 0.19.13 might be quite a serious issue

[ritchie46]

Yes, let's revert this and fix first last.