Drop the thumbnail auto-selection feature for datasets

[EDIT: It was pointed out below that I HAD ALREADY IMPLEMENTED THIS WITH A FEATURE FLAG some time ago! I must have turned it off in prod. at some point, under the assumption it was no longer causing issues. While I could simply close this issue in shame, quietly, I'm leaving it open at least temporarily, for a potential discussion of changing it further by reverting the behavior, and making this auto-select something an instance needs to opt in rather than out of]

This is something I want to get rid of, at least in our own prod. instance.

The code is buggy and occasionally causes serious problems. There was a recent prod. issue with a dataset with 3K TIF images neither one of which would produce a thumbnail. The draft was unloadable; presumably because the silly page would attempt to go through the entire 3K of them every time. Which implies that the thumbnail failure flag is not getting set properly. But I would argue that it is wrong to have to go through the entire list of files even once in a case like that.

I do not think that the feature is useful either. It is very easy for a dataset author to pick a specific image to serve as the dedicated dataset thumbnail. Making that decision for them, randomly, is a bit patronizing and unnecessary. I do not believe that is worth the extra code complexity behind it.

I will start with providing a feature flag for disabling it. But I would vote for just losing it completely.

Start by reviewing the existing thumbnail auto-selection feature flag and the comment history to determine whether the intended change is to revert the behavior or change its default. The issue does not name files or tests, and completion depends on resolving that design discussion before implementation can be defined.