Missing support of certain Arrow types and issues with converting supported Arrow data types to DH ones

[jmao-denver]

These problems were discovered in testing the new arrow module in Python on the supposedly supported Arrow data types. 'timestamp' round-trip doesn't work if it starts from Arrow with a tz other than UTC.

    @unittest.skip("Not correctly widened")
    def test_arrow_types_unsigned_integers(self):
        with self.subTest("unsigned integers"):
            pa_types = [
                pa.uint16(),
            ]
            pa_data = [
                pa.array([2 ** 16 - 1, 0]),
            ]

    @unittest.skip("Not correctly converted by DH")
    def test_arrow_types_time(self):
        pa_types = [
            pa.time64('ns'),
            pa.date32(),
            pa.timestamp('ns', tz='Europe/Paris'),
        ]

        pa_data = [
            pa.array([1_000_001, 1_000_002]),
            pa.array([datetime(2022, 12, 7), datetime(2022, 12, 30)]),
            pa.array([pd.Timestamp('2017-01-01T12:01:01', tz='UTC'),
                      pd.Timestamp('2017-01-01T11:01:01', tz='Europe/Paris')]),
        ]
        self.verify_type_conversion(pa_types=pa_types, pa_data=pa_data)

Also the following Arrow types appear to be reasonable to support

            pa.null(),
            pa.uint8(),
            pa.uint32(),
            pa.uint64(),
            pa.float16(),
            pa.time32('s'),
            pa.time32('ms'),
            pa.time64('us'),
            pa.timestamp('s', tz=None),
            pa.timestamp('ms', tz=None),
            pa.timestamp('us', tz=None),
            pa.duration('s'),
            pa.duration('ms'),
            pa.duration('us'),
            pa.duration('ns'),
            pa.month_day_nano_interval(),
            pa.binary(),
            pa.large_binary(),
            pa.large_string(),
            pa.large_utf8(),

[jmao-denver]

from datetime import datetime
from typing import List, Any

import numpy as np
import pandas as pd
import pyarrow as pa
from deephaven import arrow as dharrow, dtypes, new_table, time_table

def verify_type_conversion(self, pa_types: List[pa.DataType], pa_data: List[Any]):
    fields = [pa.field(f"f{i}", ty) for i, ty in enumerate(pa_types)]
    schema = pa.schema(fields)
    pa_table = pa.table(pa_data, schema=schema)
    dh_table = dharrow.to_table(pa_table)
    arrow_table = dharrow.to_arrow(dh_table)
    print(pa_table.equals(arrow_table))

pa_types = [
    pa.time64('ns'),
    pa.date32(),
    pa.timestamp('ns', tz='Europe/Paris'),
]

pa_data = [
    pa.array([1_000_001, 1_000_002]),
    pa.array([datetime(2022, 12, 7), datetime(2022, 12, 30)]),
    pa.array([pd.Timestamp('2017-01-01T12:01:01', tz='UTC'),
              pd.Timestamp('2017-01-01T11:01:01', tz='Europe/Paris')]),
]
verify_type_conversion(pa_types=pa_types, pa_data=pa_data)

This should work in the WebUI's Python console with the PR #3216 checked out.

[chipkent]

Should support timestamp with and without a timezone specified.