Maybe that's why they're building Fuchsia.

Wouldn't really help -- the biggest result would be that vendors would have less of a reason to release things under the GPL.

It could help, depending on one's goals - one of the most notorious issues (or features, depending on who you ask) with the Linux development process is ruthless interface refactoring which requires device drivers and board support packages to be in mainline or to become stale almost instantly.

A smaller surface area with a stable HAL would separate the proprietary / often bad board support code from the kernel itself and allow core changes to progress without requiring software-adverse hardware manufacturers to keep up.

Whether or not this is a good thing is certainly debatable, but I'm not convinced the Linux "get to mainline or get wrecked" approach has stood the test of time well.

That's a non-issue because a microkernel system call interface is stable. Drivers will simply continue to work in perpetuity. That seems like a huge help to me.

Why would you assume that? The set of required messages for drivers will increase over time, adding power saving commands, new hooks for higher level Android bits, and so on.

"Your WiFi only supports Android Kernel API 3.9. You need 7.0 to run this version of Google play".

The Linux kernel already has a stable user space ABI. If your hypothesis was valid, you would be able to upgrade the Android userspace on an old kernel with no problem. But you can't -- and that's with just one well specified ABI with a standardized, portable subset to worry about.