Some FLE tests run transactions but they don't wrap them a retry loop to retry transient error (like stale routers).  SERVER-118778 spotted a case for find_and_modify.js but txn_update_range.js seem to have the same problem.
At this point, we should make a through research and ensure all the fle tests wrap startTransaction within a retry loop.

This only effects master given the multi-router has introduced this case, before no router could be stale.