This PR fixes an issue where, after the initial state in a state machine test has been simplified, it will not be re-complicated when this results in a successful run. As a result, when this situation is encountered, the test will report a "minimal failing input" that is actually a successful input - the input that is just barely simpler than the minimal failing input.
I've also added a proptest that detects the failure, along with two regression seeds, for good measure.
This PR fixes an issue where, after the initial state in a state machine test has been simplified, it will not be re-complicated when this results in a successful run. As a result, when this situation is encountered, the test will report a "minimal failing input" that is actually a successful input - the input that is just barely simpler than the minimal failing input.
I've also added a proptest that detects the failure, along with two regression seeds, for good measure.