Revert tooltip observer change as it does not work properly.

[JDutil]

fixes #2864

The change introduced by https://github.com/solidusio/solidus/pull/2421/files caused a regression in tooltips where they don't go away when the element is removed from the DOM. It appears the MutationObserver is meant for detecting child elements being removed from the target element, and it does not actually observe the target element itself being removed.
https://developer.mozilla.org/en-US/docs/Web/API/MutationObserver
https://developers.google.com/web/updates/2012/02/Detect-DOM-changes-with-Mutation-Observers
https://stackoverflow.com/questions/28598340/mutation-observer-fails-to-detect-elements-dom-removal

So we can either revert the change to using MutationObserver or we would need to change it to observe a different target element, such as, the parent element of our actual target. I gave a go at wrapping the the target with a parent node, and observing the parent node and I couldn't easily get it to work either. So reverting seems easiest solution for now, and I changed from tooltip.hide() to tooltip.dispose() to unbind events as well.

[kennyadsl]

Yes the (hacky) tricky here was intercepting the aria-describedby attribute change on the element itself, which seemed to happen on element removal as well on Chrome and Safari, but it's (now?) not happening on Firefox.

Wrapping into a parent div is another hack, not better than the timeout solution in my opinion, so 👍 with reverting that commit.

[ericsaupe]

Good revert. Thanks