I can see it would be a useful feature and issue already raised for it, but heres some background whereby I dont see it as critical.
I originally developed Jaikoz Music Tagger - this was a more traditional tagger (althoug it still had some movel features), and it allowed you to load your files into a spreadsheet like view and then match your songs to MusicBrainz/Disocgs or do manual edits. All changes were visibles on screen but just held in memory until you actually elected to Save. But I found that even though users could check the results before saving they seldom did, instead they would save without checking and then complain about the changes made afterwards.
So with SongKong we save as we go along, but all changes are recorded in database. So after SongKong has run if you find problems afterwards you can use the Undo Fixes task to revert the changes when you find issues, and this can be applied at folder level no need to revert everything.
Now in your case you have some workflows you could consider
Run with Preview enabled, check the results if happy run again without Preview, the results are going to be 99% the same, not guranteed to be identical (changes in online database since first run, timnng issues etc) but in most cases they will exactly the same. If you do hit issues with an album then there is Undo Fixes. But with such a large music collection how could you realistically check it all anyway?
Alternative, work on a smaller dataset e.g bands starting with A, then can check more thorougly before committing. It is unlikely that SongKong is going to be matching the wrong album ( the worst case is maybe not the preferred version when there are multiple versions that a match, or boxset vs orginal albums). It is more the case that you dont like the way a particular tag is formatted (e.g multiple artists query) so once ironed out on a smaller subset options should remain correct on collection as a whole.