Combine Python Flink load test (batch/stream) no parallelism

One of the Combine load test cases, which involves a global combiner and data stream of 200M elements, takes too long on Flink. Flink is able to process only a half of that data stream within 1 hour, which is too long for a Jenkins job.

Job's definition: https://github.com/apache/beam/blob/master/.test-infra/jenkins/job_LoadTests_Combine_Flink_Python.groovy#L36

Test pipeline: https://github.com/apache/beam/blob/master/sdks/python/apache_beam/testing/load_tests/combine_test.py

Imported from Jira [BEAM-10852](https://issues.apache.org/jira/browse/BEAM-10852). Original Jira may contain additional context.
Reported by: kamilwu.