You hit the nail on its head. Google usually discontinues products which have high maintenance cost so even if it turns out that YT Premium makes less money than people watching ads (and I highly doubt the average person watches more than $10 worth of ads in a month), they'll still keep it.
Not really, for example Reader had really low maintenance costs. In fact at some point it was maintained by some of its former developers in their 20% time with no official head count.
I think it is a side effect of what keeps google from turning into IBM. Ossification is a perennial problem for large corporations especially traditional growth companies. They make their few big things then iterate on them. The "ADHD" shown by google, while it is annoying to users to have a perfectly serviceable piece of software discontinued unceremoniously, probably keeps the internal structure from becoming rigid and failing to adapt. I think they are probably slightly too far over on the side of kill stuff and move on but I'm not a multibillion dollar company.
Having zero headcount is the death knell for a project about to be cancelled. It only had low maintenance costs because the team was disbanded, presumably because it wasn't worth paying to maintain.
The story behind reader is more complicated than that.