Deaths are recorded when someone is declared dead. A death record doesn't necessarily mean the person has been found, though the different last name on the death record is odd...
I appreciate a death record is made when someone is declared dead. But the existence of a record does not mean it is accessible to a researcher unless they know where to look. In 2006 accessing a new death record may have meant chasing paper records (or microfiche) and may even have required some knowledge such as place and time of death to narrow the search. But since those days many more records are online and searches can be made on much less information. Hence it is quite possible that a death could have occurred in 2006 (or been recorded due to a declaration) but not been found by a researcher for some years. Today a researcher would expect to find a record more quickly.
I take the point that a death record could be due to a declaration of death rather than a person being found. However, in this case, the minimum that the Social Security record supports is that she was alive after her disappearance in 1971, since it records the name changes and the change to O'Neil is recorded as 1976 (3 years after a supporting marriage record to Leslie O'neil). I have also learnt my lesson that it is worth noting all records in my posts (!) as there is also an entry in the US Public Records Index showing Mary Ann Oneil living in Florida.
I remain as surprised and cautious as others about this case and developments, but cannot see a good alternative explanation for the Social Security record than that she left voluntarily. The other records are just supporting and no detail can be wholly relied upon. But ever since the Horrell case I have accepted that such cases do happen. What is very odd is that LE retain an interest as you would expect them to check government records as a first step. Either they have not done so or there is something wrong with the record.