I work on the Smart Lock team at Google and we debated the wording of this dialog for months and extensively studied user interaction very carefully in the lab and have sampled millions of user actions on the dialog in live apps and websites to monitor save, decline, and cancel rates. Note that the Never selection is only an opt-out for the app in question and does not affect other unrelated apps.
A few of the findings from user research and feedback:
When asking users in the lab what they might do if they did not want to select either option (Never or Save password), most noted that they would tap outside or tap the back button, which are the established cancel actions in Android UI design patterns. So we did not find an issue with discoverability of the cancel action if strongly preferred.
Adding a third option to the dialog significantly complicates the dialog (both visually and cognitively) and adds substantial confusion and indecision for the user, in that they have to process the difference between Never and No thanks, rather than making a simpler decision between two options.
In practice, we found that user preference to save a password very rarely changed: if user declined once, they would overwhelming choose to decline again if prompted again later, even after some time. In practice, this means that a No thanks (not now) option presented the worst possible outcome: user does not get into the saved credential state (with auto sign-in) and was also repeatedly prompted and interrupted whenever they sign in to the app again.
Thus, the current implementation only presents the two options (Never / Save password, or Save account for non-password credentials) to steer the user toward a decision putting them in a better state, where they are not prompted again for this app, or opt in to benefit from saved information. From the data we've see, the benefit of not continually prompting a user in the common decline case greatly outweighs the lost opportunity to save easily after having declined previously.
But the question does raise a good point that removing an app from the "Never save" list is a lot of steps (it can be done in Chrome settings, passwords.google.com, or Android Google Settings) and might be worth revisiting, especially in the event it was accidentally selected, as noted in question. And if some change in user affinity for saving password data is detected (e.g. user starts opting in to save data for other apps), it might be appropriate to prompt the user to review the apps for which they've opted out.
(Note: one other variation we considered was having "Never save" apply to just the username in question, but that also added a lot of complexity and confusion, and in the studies, the majority intent from users in declining was not to save any password information with Google at all for the app, it was rare that user would have multiple accounts and only want to save one, but not another)
