Investigate: does BigQueryIO honor write disposition in all cases?

I am not certain that we have proper testing of the spread of write dispositions, and I do not see enough (any?) testing of the non-default values. We should verify that these are properly used, even for large multi-table copies, etc.

Imported from Jira [BEAM-1181](https://issues.apache.org/jira/browse/BEAM-1181). Original Jira may contain additional context.
Reported by: dhalperi.

Start by locating BigQueryIO and its existing write-disposition tests. Check whether non-default dispositions are covered, including large multi-table copies, and verify that each relevant case behaves as expected. The work is done when the disposition behavior is tested across the identified cases.

Written by the indexing model from the issue text.