BUG: Data corruption when mode is performed on timedelta64 dtype in pandas-2.0

[galipremsagar]

Pandas version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

In [22]: df = pd.DataFrame(
    ...:             {
    ...:                 "a": ["hello", "world", "pandas", "numpy", "df"],
    ...:                 "b": pd.Series(
    ...:                     np.array([1, 21, 21, 11, 11],
    ...:                     dtype="timedelta64[s]"),
    ...:                     index=["a", "b", "c", "d", " e"],
    ...:                 ),
    ...:             },
    ...:             index=["a", "b", "c", "d", " e"],
    ...:         )

In [23]: df
Out[23]: 
         a               b
a    hello 0 days 00:00:01
b    world 0 days 00:00:21
c   pandas 0 days 00:00:21
d    numpy 0 days 00:00:11
 e      df 0 days 00:00:11

In [24]: df.mode()
Out[24]: 
        a                    b
0      df      0 days 00:00:11
1   hello      0 days 00:00:21
2   numpy 106751 days 23:47:15 <--  # Expected NaT
3  pandas 106751 days 23:47:15 <--  # Expected NaT
4   world 106751 days 23:47:15 <--  # Expected NaT

Issue Description

In pandas-2.0, when we perform a mode operation on timedelta64 dtype, it looks like NaT values are not being returned correctly. For example see the pandas-1.5.x behavior below:

In [27]: df = pd.DataFrame(
    ...:             {
    ...:                 "a": ["hello", "world", "pandas", "numpy", "df"],
    ...:                 "b": pd.Series(
    ...:                     [1, 21, 21, 11, 11],
    ...:                     dtype="timedelta64[ns]",
    ...:                     index=["a", "b", "c", "d", " e"],
    ...:                 ),
    ...:             },
    ...:             index=["a", "b", "c", "d", " e"],
    ...:         )

In [28]: df.mode()
Out[28]: 
        a                         b
0      df 0 days 00:00:00.000000011
1   hello 0 days 00:00:00.000000021
2   numpy                       NaT
3  pandas                       NaT
4   world                       NaT

Expected Behavior

In [24]: df.mode()
Out[24]: 
        a                    b
0      df      0 days 00:00:11
1   hello      0 days 00:00:21
2   numpy                  NaT
3  pandas                  NaT
4   world                  NaT

Installed Versions

INSTALLED VERSIONS

commit : 965ceca
python : 3.10.11.final.0
python-bits : 64
OS : Linux
OS-release : 4.15.0-76-generic
Version : #86-Ubuntu SMP Fri Jan 17 17:24:28 UTC 2020
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 2.0.2
numpy : 1.24.3
pytz : 2023.3
dateutil : 2.8.2
setuptools : 67.7.2
pip : 23.1.2
Cython : 0.29.35
pytest : 7.3.1
hypothesis : 6.75.7
sphinx : 5.3.0
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 3.1.2
IPython : 8.13.2
pandas_datareader: None
bs4 : 4.12.2
bottleneck : None
brotli :
fastparquet : None
fsspec : 2023.5.0
gcsfs : None
matplotlib : None
numba : 0.57.0
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : 11.0.0
pyreadstat : None
pyxlsb : None
s3fs : 2023.5.0
scipy : 1.10.1
snappy :
sqlalchemy : 2.0.15
tables : None
tabulate : 0.9.0
xarray : None
xlrd : None
zstandard : None
tzdata : 2023.3
qtpy : None
pyqt5 : None

[asishm-wk]

Simpler repro:

import pandas as pd
df = pd.DataFrame(
    {
        '0': pd.Series(['a', 'b', 'c']),
        '1': pd.Series(
            np.array([1, 21],
            dtype="timedelta64[s]"),
        )
    }
)
df
# outputs
   0                    1
0  a      0 days 00:00:01
1  b      0 days 00:00:21
2  c 106751 days 23:47:15