But a de facto standard (ie one defined by a popular implementation) forces sites to be compatible or be ignored. More monolithic standard => stronger coding conventions => more robust web. Or am I being totally idealistic?
Robustness often comes from the opposite approach. It's the willingness to throw out ideas and code that are observed to be broken. Replacing JavaScript with a more sensible scripting language like Lua or Python would increase the robustness of the web, while standardizing it further only entrenches the problem deeper.
I call on your "no need to rehash it here." Without even one good example, I'd say you're just trying to be elitist.
Let me help set you on the right path. Google for "javascript semicolon insertion". After reading about it, it should be very apparent to you how it's a very large flaw with the language, and how it can be extremely harmful.
Do some research into JavaScript's equality operators. Any good programmer will be sick to his or her stomach after reading about how broken such basic, yet essential, functionality like that is.
And that's just the start, my friend. I encourage you to do further research on your own. You'll find no shortage of serious issues to learn more about.
No, it's totally the opposite. They only have to be "compatible" with this one implementation, and are free to rely on all implementation-specific bugs or implementation defined behavior when doing so.
Think IE box model around the IE5/6 age. Or hell, supporting Office Documents. Even if you'd have the source, being compatible certainly wouldn't have gotten much easier.
Note that WebKit is open source and Opera is having problems being bug-compatible with it.
Although there is a chance here for the -prefix-rabbit-hole to yawn wide, there's also an opportunity for a (nearly-)universally-supported rendering engine to emerge that could, one day, get into bed with the standards committee. Surely this would be a good endgame for the browser wars?
Note that coding to an implementation rather than a standard does stifle innovation, because when you code to target the implementation you inevitably rely on bugs in the implementation, without ever knowing or trying to. From the perspective of an implementor, would you fix a bug even if it meant breaking an unknowably large number of websites? If not, you've adopted that bug for the rest of time, and every subsequent implementation, ever, will be forced to implement the bug as well when they could be busy implementing new features. Now compound this by thousands of bugs, and reflect on all the time spent implementing Quirks Mode in every browser engine that exists today.
I do agree with you that in a perfect world we wouldn't need multiple implementations, because bugs would not exist and the lone implementation would track the standard perfectly, and because the standard would be controlled by neutral third-parties. But I think it's much too soon to declare the current era the browser endgame.