Fix: pandas NaT in datetime/timedelta columns stored as garbage values

[mdbenito]

Currently NaT/None in datetime64 and timedelta64 DataFrame columns are stored as the sentinel value (INT64_MIN ~= 1677-09-21) instead of NULL.

This PR addresses that.

To reproduce before the PR:

import datetime
import numpy as np
import pandas as pd
import ladybug

db = ladybug.Database()
conn = ladybug.Connection(db)

# datetime NaT
conn.execute("CREATE NODE TABLE t_dt (id INT64, ts TIMESTAMP, PRIMARY KEY (id))")

nat = np.datetime64("NaT", "ns")
df_dt = pd.DataFrame(
    {
        "id": [1, 2],
        "ts": np.array([np.datetime64("2024-01-15"), nat], dtype="datetime64[ns]"),
    }
)
conn.execute(
    "COPY t_dt FROM (LOAD FROM $df RETURN CAST(id AS INT64) AS id, CAST(ts AS TIMESTAMP) AS ts)",
    {"df": df_dt},
)

for row in (
    conn.execute("MATCH (t:t_dt) RETURN t.id, t.ts ORDER BY t.id")
    .get_as_df()
    .itertuples(index=False)
):
    print(f"id={row[0]:<2}  ts={row[1]!r}")

# timedelta NaT
conn.execute("CREATE NODE TABLE t_td (id INT64, dur INTERVAL, PRIMARY KEY (id))")

nat_td = np.timedelta64("NaT", "ns")
df_td = pd.DataFrame(
    {
        "id": [1, 2],
        "td": np.array(
            [np.timedelta64(3600000000000, "ns"), nat_td], dtype="timedelta64[ns]"
        ),
    }
)
conn.execute(
    "COPY t_td FROM (LOAD FROM $df RETURN CAST(id AS INT64) AS id, CAST(td AS INTERVAL) AS dur)",
    {"df": df_td},
)

for row in (
    conn.execute("MATCH (t:t_td) RETURN t.id, t.dur ORDER BY t.id")
    .get_as_df()
    .itertuples(index=False)
):
    print(f"id={row[0]:<2}  dur={row[1]!r}")

Running that script yields:

id=1   ts=Timestamp('2024-01-15 00:00:00')
id=2   ts=Timestamp('1677-09-21 00:12:43.145225')
id=1   dur=Timedelta('0 days 01:00:00')
id=2   dur=Timedelta('-106752 days +00:12:43.145225')

And after the fix:

id=1   ts=Timestamp('2024-01-15 00:00:00')
id=2   ts=NaT
id=1   dur=Timedelta('0 days 01:00:00')
id=2   dur=NaT