map_elements doesn't retain the datetime's timezone

[yiteng-guo]

Checks

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

Reproducible example

> dt = pd.Timestamp.now(tz="America/New_York").to_pydatetime()

> pl.select(x=dt).select(pl.col.x.map_elements(lambda x: x, pl.Datetime(time_zone="America/New_York")))

SchemaError: expected output type 'Datetime(Microseconds, Some("America/New_York"))', got 'Datetime(Microseconds, None)'; set `return_dtype` to the proper datatype

> pl.select(x=dt).select(pl.col.x.map_elements(lambda x: x, pl.Datetime("ns", time_zone="America/New_York")))

SchemaError: expected output type 'Datetime(Nanoseconds, Some("America/New_York"))', got 'Datetime(Microseconds, None)'; set `return_dtype` to the proper datatype

> pl.select(x=pl.lit(dt, dtype=pl.Datetime("ns", time_zone="America/New_York")))

> pl.select(x=pl.lit(pd.Timestamp.now(), dtype=pl.Datetime("ns", time_zone="America/New_York"))

Log output

No response

Issue description

I think there're two bugs/issues in this example

• polars always treats the datetime return of map_elements as a Datetime(Microseconds, None) even if a timezone exists for the return datetime.
• polars dtype conversion in map_elements is stricter than pl.lit. This leads to issues where pl.Dateime("ns") can't be used in return_dtype when a datetime is returned. I think this should have the same dtype rule as pl.lit.

Expected behavior

All examples above should return a dataframe with the specified dtype w/o any error.

Installed versions

``` In [17]: pl.show_versions() --------Version info--------- Polars: 1.9.0 Index type: UInt32 Python: 3.10.13 ```

[cmdlineluser]

@MarcoGorelli I think this may fall under the A-timeseries label when you get a chance.