feat: convert fixed-offset timezones to respective Etc timezone from time zone database

[MarcoGorelli]

closes #12893

demo with the example reported in the issue:

tz = dt.timezone(dt.timedelta(hours=10))
df = pd.DataFrame.from_records(
    data=[(dt.datetime(2024, 1, 1, tzinfo=tz),)], columns=("naughty_date",)
)
print(df)
df.to_parquet("naughty.parquet")
print(pl.read_parquet("naughty.parquet"))

tz = dt.timezone(dt.timedelta(hours=-10))
df = pd.DataFrame.from_records(
    data=[(dt.datetime(2024, 1, 1, tzinfo=tz),)], columns=("naughty_date",)
)
print(df)
df.to_parquet("naughty.parquet")
print(pl.read_parquet("naughty.parquet"))

               naughty_date
0 2024-01-01 00:00:00+10:00
shape: (1, 1)
┌──────────────────────────┐
│ naughty_date             │
│ ---                      │
│ datetime[ns, Etc/GMT-10] │
╞══════════════════════════╡
│ 2024-01-01 00:00:00 +10  │
└──────────────────────────┘
               naughty_date
0 2024-01-01 00:00:00-10:00
shape: (1, 1)
┌──────────────────────────┐
│ naughty_date             │
│ ---                      │
│ datetime[ns, Etc/GMT+10] │
╞══════════════════════════╡
│ 2024-01-01 00:00:00 -10  │
└──────────────────────────┘

[ritchie46]

Maybe make this one static. Doesn't have to be inlined everywhere we use it I would say. (or aren't we going to use it more often?)

[stinodego]

Does/should this also work for parsing regular time zone inputs, e.g. Datetime(time_zone="+10:00")? And csv files? Just wondering.

[MarcoGorelli]

Does/should this also work for parsing regular time zone inputs, e.g. Datetime(time_zone="+10:00")? And csv files? Just wondering.

For Datetime(time=...) it doesn't at the moment, no - not sure that these should be encouraged too much to be honest.
OK for I/O where the user might not have control over the dtype, but if they're instantiating a Datetime object then they can directly set the equivalent timezone-database time zone

For csv files, they don't store dtypes anyway, so it's all good (and offset-aware inputs are converted to UTC, as they are in to_datetime, and as pyarrow also does)

[stinodego]

Right, read_csv can already read strings like 2024-02-28T00:00:00.000000+0800. Great - merging then!