The main pro of my proposal is safeguarding the user from losing funds.
The other proposal fails to do so, as it still allows the user to cancel a tx (which was to be completed by another party) without sweeping its input(s), and then be subjected to a play attack within a limited time window.
[EDIT:]
This risk of losing funds is easily avoided though as noted below in Play Attacks and possible mitigations
The other pro of my solution is preserving tx monotonicity, as argued for in Replay Attacks and possible mitigations