If a bulk write undergoes a retryable write, we select the new server here.

However, the bulk operation loop is not made aware of that newly selected server, and reuses the original server id. If that server has stepped down, than it's likely we'll fail the subsequent commands for the same reason (causing another retryable write).