[ML-49316] Support MonthMid and MonthEnd for DeepAR

[Lanz-db]

This PR fixes the bug that when user dataset has monthly frequency and the day of the month is not the first day, DeepAR will fail. The bug results from this line,

new_index_full = pd.date_range(total_min, total_max, freq=frequency)

freq is "MS" so the generated new_index_full will always be the first day of month. So this line,

df.reindex(new_index_full)

will generate a df with all rows in target column to be NaN.

To fix the bug, this PR introduces

1. a helper function, validate_and_generate_index , to generate a complete time index for the given DataFrame based on the specified frequency. If it is monthly frequency, it will generate the index based on the given day of month, also detect if it is the end of month.
2. Use this helper function instead of pd.date_range(total_min, total_max, freq=frequency)

To test the function,

• Unit tests added in utils_test.py
  Run the below command locally

PYTHONPATH=~/automl/runtime/ pytest tests/automl_runtime/forecast/deepar/utils_test.py

• manual e2e test
  Month start: notebook
  Month mid: notebook
  Month end: notebook