BUG: groupby(Grouper) with all-NaT grouping keys

[jbrockmendel]

• 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 master branch of pandas.

Reproducible Example

import pandas as pd

df = pd.DataFrame({"date": [pd.NaT], "value": [11]})
grouper = pd.Grouper(freq="M", key="date")

>>> gb = df.groupby(grouper)
[...]
  File "pandas/core/resample.py", line 1878, in _get_timestamp_range_edges
    first = first.normalize()
AttributeError: 'NaTType' object has no attribute 'normalize'

### Issue Description

Example derived from test_timegrouper_apply_return_type_series


### Expected Behavior

Either not raise or raise at construction of the Grouper object with a useful exception message.

### Installed Versions

master

[WKKO]

Sorry to bother you. Do you know the solution to this problem? I've been in trouble recently

[burnpanck]

I believe this is a manifestation of #24983: With the current design, pd.NaT isn't a timestamp at all; it ambiguously refers to either an undefined timedelta or an undefined timestamp. You might get lucky by explicitly making the "date" column a timestamp, but the grouper might nonetheless fail once it reaches the nat time.

[jamsden]

Any workaround identified? i.e., could I test for all NaT values and return an empty Series?

[kfchou]

Since it wouldn't make sense to count or add anything in this situation, here's my workaround:

    if all(df['key'].isna()):
        df['key'] = pd.DataFrame(columns=df.columns)