It's silly to depend on an email for authentication, then allow the hacker to just delete the email address before they change the password. Giving the old address the right of first refusal defeats that kind of attack and should be dead simple to implement since the framework was already laid down for the "verify your email" step during setup.
I don't think so. Why, in your scenario, would they file a help request saying the account had been compromised? They might file a request with some other content, but not that.
Your general point is valid, but I think the OP has probably figured out a set of features from which one could pretty reliably tell that something was amiss. And all he's suggesting is that such cases get bounced up to a human.
Google and other service providers do have data to evaluate the benefit and cost of making decisions based on patterns, and they probably do.
This series of events could easily occur in legitimate cases. Say you lose or destroy your cellphone. Since you only ever logged in via your phone you don't know the password. Your recovery email was attached to a service you don't use because you normally use gmail. I'm not saying this scenario is a good idea just that it's probably quite common.
As a software developer I often hear from well meaning users that are appalled that software didn't do-the-right-thing in some complex scenario that appears to have an obvious solution because the desired outcome in obvious. In reality, handling the corner cases is complex. Adding these obvious solutions to the code easily leads to even worse situations.