perf(rust, python): parse time zones outside of downcast_iter() in replace_time_zone

[MarcoGorelli]

Noticed while looking into #10090

I probably introduced this inefficiency the first time round, sorry about that

[orlp]

I think this can be written quite a bit cleaner than the current nested match (didn't test it):

fn parse_timezone(s: &str) -> PolarsResult<Tz> {
    s.parse().map_err(|e| polars_err!(ComputeError: "unable to parse time zone '{s}': {e}"))
}

pub fn replace_time_zone(
    ca: &DatetimeChunked,
    time_zone: Option<&str>,
    use_earliest: Option<bool>,
) -> PolarsResult<DatetimeChunked> {
    let from = parse_timezone(ca.time_zone().as_deref().unwrap_or("UTC"))?;
    let to = parse_timezone(time_zone.unwrap_or("UTC"))?;
    let chunks = ca.downcast_iter().map(|arr| {
        replace_time_zone_kernel(arr, ca.time_unit().to_arrow(), &from, &to, use_earliest)
    });
    let ca = ChunkedArray::try_from_chunk_iter(ca.name(), chunks)?;
    let mut out = ca.into_datetime(ca.time_unit(), time_zone.map(|x| x.to_string()));
    out.set_sorted_flag(ca.is_sorted_flag());
    Ok(out)
}

The question mark is your friend :)

[MarcoGorelli]

much cleaner, thanks 🙏