When debugging why the ordered write timestamp constraint wasn't firing, I'm observing the following:

• __wt_update_serial initializes prev_upd_ts to the 0 timestamp
• The prev_upd_ts is assigned in calls to __wt_txn_update_check as an out-parameter.
• However, if the compare and swap succeeds on the first try, we never call _wt_txn_update_check.
• This overrides a previously correct value for the 0 timestamp that trivially succeeds an ordered check.