Avoid pandas resample crash in Windows Python 3.12 forecast tests

[Copilot]

The build (windows-latest, 3.12) job was failing with a fatal access violation during forecast test setup. The crash came from pandas daily resampling in shared test dataset preprocessing.

• Root cause

  • Forecast test fixtures in test_extra_models.py and test_forecast.py normalized hourly data with:

    df = df.set_index("timeStamp")
    df = df.resample("D").mean()

  • On Windows + Python 3.12, that resample path crashes inside pandas.

• Change

  • Replace resample-based daily aggregation with explicit day bucketing plus grouped mean:

    df["timeStamp"] = pd.to_datetime(df["timeStamp"]).dt.floor("D")
    df = df.groupby("timeStamp", as_index=False).mean(numeric_only=True)

  • Keep the rest of the fixture behavior unchanged (ffill, trailing NaN row removal).

• Coverage

  • Add a focused regression test in test_extra_models.py that patches pandas.read_csv and verifies the helper still produces the expected daily aggregates without using resample.

• Effect

  • Preserves test intent and dataset shape while avoiding the Windows-specific pandas crash path in the failing job.