CDRIVER-3268 discovered a bug in how the collection's unacknowledged write concern resulted in an unnecessary error within a transaction (though the collection's write concern would not be applied in that case).

The test that caught it was a spec test introduced in CDRIVER-3089. But those spec tests are only for insert and bulk write. In addition, the spec tests do not test read concern/write concern/read preferences at the operation level (because not all drivers support it).

I think we can try extending the comprehensive test-mongoc-opts.c to get more test coverage and uncover any other potential bugs. We can add to the matrix testing how read preferences, write concern, and read concern are applied as an operation in the middle of a transaction, and how the session default options and transaction options are handled.