Running patch builds have exposed a number of issues with this test:

1. Chunk migrations need to be allowed to fail with an Interrupted error code, since we will be interrupting them whenever we run a multi update, this is similar to the test for refine shard key.
2. Each thread needs to call getClusterParameter for pauseMigrationsDuringMultiUpdates as part of init to ensure that its mongos has refreshed after we set it.
3. Update and delete commands should include a number of additional (irrelevant) updates/deletes in their batches, and these batch arrays should be shuffled before sending. This is to ensure that multi updates are properly coordinated regardless of where they appear in a batch.