Weekly column_anomalies fail with python error on postgres on 0.26.0 #2370

Describe the bug
column_anomalies fails with '<' not supported between 'NoneType' and 'str' on weekly buckets since 0.26.0 (fine on 0.25.1).

'<' not supported between instances of 'NoneType' and 'str'
    
in macro store_anomaly_test_results (macros/edr/data_monitoring/anomaly_detection/store_anomaly_test_results.sql)
called by macro test_column_anomalies (macros/edr/tests/test_column_anomalies.sql)
called by test elementary_column_<snip> (<snip>)

To Reproduce

• Have a weekly time_bucket with a timestamp_column.
• Have a column_anomalies test
• Data to have a row dated on the last Sunday of the most recent complete week
• Data to have enough history for that row to get an anomaly score

Example config:

    config:
      elementary:
        timestamp_column: full_date
        time_bucket:
          period: week
          count: 1

Expected behavior
The test should not fail with python error raised from elementary core.

Screenshots
n/a

Environment (please complete the following information):
elementary dbt package 0.26.0
edr 0.26.0
dbt-core 1.12.5, dbt-postgres
Postgres

Additional context

AI gives the following summary:

Precursor:

On Postgres, get_start_bucket_in_data calls dbt.datediff(..., 'week'), and that function starts a new week on Sunday (it compares dow, where Sunday is 0). Elementary's week buckets run Monday to Monday, so Sunday rows go into the following week.

The last Sunday before max(edr_bucket_end) maps to a bucket that doesn't exist, and the left join buckets in column_monitoring_query returns a metric row with NULL bucket_start and bucket_end. That row still gets an anomaly score.

This bug predates 0.26.0; every weekly sum has been shifted by one day.

Trigger:

0.26.0 added | sort(attribute='bucket_end') in get_anomaly_test_result_row. Sorting that NULL-bucket row against the string dates throws the error above.

(Monthly and daily buckets aren't affected.)