The contextual approach we had going for a while makes the most sense to me and many other users. I don't understand why it was scrapped but I won't speculate on it... I'll only say that it satisfied both camps and with no obvious intrusion or loss of functionality.
What you seem to fail to realize is that the right approach will look like a "pick one" to users depending on their personal usage. Touch the screen with your narrow stylus and the UI gives you stylus-based input modes. Touch it with your big greasy finger and you get big fat icons and gestures. Simple. Elegant. Effective. Win-win.