BigQuery: `last_modified` underreports freshness for streamed sources

Just an observation from our implementation

**Issue summary:**
On BigQuery, the adapter derives freshness from `Table.modified` in `clients/dbt_state/src/dbt_state/adapters/bigquery.py`. For tables fed by streaming ingestion (`tabledata.insertAll` or the Storage Write API), `Table.modified` doesn't advance per streamed row — it moves on buffer flush, load jobs, or DML.

This means that the state adapter may reuse models when new rows are available in a source's streaming buffer (in our case this occurs, it may be circumventable on the DE side by flushing more often).