Increase wait time between polls when not running

Dataproc serverless batch jobs take at least 60 seconds to
start up, which leads to a lot of meaningless polling, and
increases the chance of getting a transient error.
In the absolute worst case this will add 9 seconds to
a model's runtime, but I would be very surprised if
people are using Python models to do less than
10 seconds of processing, and doubt there will be
any real world impact by this change

dbt/adapters/bigquery/dataproc/batch.py

@@ -35,7 +35,7 @@ def poll_batch_job(
     response = None
     run_time = 0
     while state in _BATCH_RUNNING_STATES and run_time < timeout:
-        time.sleep(1)
+        time.sleep(1) if state == Batch.State.RUNNING else time.sleep(10)
         response = job_client.get_batch(  # type: ignore
             request=GetBatchRequest(name=batch_name),  # type: ignore
             retry=Retry()