fix(rust, python): implement missing conversion to python time object

[alexander-beedie]

The AnyValue::Time match implementation for into_py was returning the underlying integer representation; this PR ensures that we now create proper python time objects instead.

Was most apparent in calls to DataFrame.rows() for any frame containing time cols. (Interestingly Series.to_list() was fine, as the Wrap implementation for TimeChunked handles it correctly).

Example

from datetime import datetime
import polars as pl

df = pl.DataFrame({
    "dtm": [
        datetime(2020,1,2,10,30,45),
        datetime(2022,3,4,23,59,59),
    ]
}).with_column(
    pl.col("dtm").cast(pl.Time).alias("tm")
)
# ┌─────────────────────┬──────────┐
# │ dtm                 ┆ tm       │
# │ ---                 ┆ ---      │
# │ datetime[μs]        ┆ time     │
# ╞═════════════════════╪══════════╡
# │ 2020-01-02 10:30:45 ┆ 10:30:45 │
# ├╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌┼╌╌╌╌╌╌╌╌╌╌┤
# │ 2022-03-04 23:59:59 ┆ 23:59:59 │
# └─────────────────────┴──────────┘

Before: (raw integers)

df.rows()
# [(datetime(2020,1,2,10,30,45), 37845000000000),
#  (datetime(2022,3,4,23,59,59), 86399000000000),]

After: (time values)

df.rows()
# [(datetime(2020,1,2,10,30,45), time(10,30,45)), 
#  (datetime(2022,3,4,23,59,59), time(23,59,59)),]

Note on updated unit tests

Quite a lot of the unit tests weren't actually validating the result; they were validating the shape of the result (which is how this one didn't get spotted earlier). I've updated most of the temporal unit tests such that they also check the returned values as well.

For example: test_filter_time would have caught this, but was written as...

assert df.filter(pl.col("t") < pl.lit(time(11, 0))).shape[0] == 3

I've now updated it (and some others) to be more explicit, eg:

assert df.filter(pl.col("t") < pl.lit(time(11, 0))).rows() == expected_times

[ritchie46]

Great, thanks!