The caption style was added to resolve T170344 but that style has since been removed, so I don't think this is needed any more - so I think really this is a question of design and whether it makes sense to make it appear like the caption is part of the table our outside the table?

consistent appearance between the two skins

Out of curiosity, why do you think the appearance need to be consistent between the two skins? I don't see any particular problem with skins looking different (in fact the content is very different depending on the skin you use so am curious to hear your perspective on this.

If consistency is the goal this should likely be defined inside https://gerrit.wikimedia.org/g/mediawiki/core/+/bd3b3c6cd7b83268101cf8693623ceaa0bff7907/includes/ResourceLoader/SkinModule.php#158

Yes, the content looks different depending on skin (and that's okay), but my impression is that especially the wikitable styling is pretty much the same across all wikimedia deployed skins, and because the background styling is already in ResourceLoader skin module, it makes sense to also apply the background consistently. If from a design point of view it makes sense to have some kind of visual connection between table and caption (I don't know), then I'm fine with it, but it should be considered that some tables make use of completely different background colors in the entire table which would leave the caption area as the only one with the initial background - then the visual connection seems rather weak and random to me.

Minerva background color was the exception until now. I don't know which change exactly is responsible for the change from white background to --background-color-neutral-subtle, but it allows us to locally apply also our zebra class to minerva, which was not possible until now. That's nice.