I think there are unanswered logistical questions around naming large areas of sea (oceans being the extreme example - it would be deeply impractical to map those as relations).

I agree that these should ideally be mapped as areas/relations so that their rendering importance can be estimated easily.

I don’t really agree with surveyability concerns here - the same surveyability issue extends to place names and other geographical features. They are to some extent subjective and impossible to define precisely, but this is unavoidable and it doesn’t mean we shouldn’t try to map them. In coastal areas, the names of sea features can be as important as land features and I strongly feel that we shouldn’t ignore them.

I know seems messy, but the way we tend to approach this in the UK is by mapping these large features as a completely separate, simplified area unattached to the coastline (example).

This means that changes to the coastline aren’t going to affect these objects and they’re much less likely to cause unexpected performance issues. People who aren’t concerned with them can completely ignore them. It makes very little difference to renderers (in fact simpler polygons are likely faster to operate on).

I’d recommend considering this approach for larger marine features in excess of a few tens of square kilometers.