Fixed a scenario where extremely large JSON rows (1gb+) being read did not always flush the writer often enough to prevent running out of memory for new allocations.