Two-steps caret movement when attribute starts at the beginning of block

[oskarwrobel]

Follow-up #4269.

Caret does not "leave" attribute element while moving left when text starts at the beginning of block element.

<p><a href="">{}foo</a></p>

Pressing left arrow does not move selection before element <a/>.

[Reinmar]

This also doesn't work when the attribute is at the end of the content.

[szymonkups]

Another issue:

1. Place selection at the end of the link.
2. Press left and right arrow keys couple times (enter and leave link using two-step caret movement).
3. Press left arrow to put selection inside link.
4. Try to exit link pressing right arrow.

Result: need to press right arrow key many times to exit the link.

[Reinmar]

I've dug a bit more into the implementation of restoreGravity() and overrideGravity() and there are a couple of things which I found at least worth checking investigation:

1. restoreGravity() should log a warning when trying to go below 0. It should also never allow going below 0. It may also happen that situations when restoreGravity() is called more times than overrideGravity() happen so often that we should simply accept this situation (and don't log).
2. It's very odd that overrideGravity() only set the custom restore callback when overrideGravityCounter == 1. What if we called overrideGravity() more than once in a row?
3. I'm afraid that the automatic restore doesn't make sense. You can test it like this: add a console log in change:range callback added by overrideGravity(). Put your caret before link and then press right, left, right, left. Now, click somewhere else in the text. Console log will be triggered 2+ times. That's because these callbacks are never removed when you press left arrow key because left arrow key is handled by bindTwoStepCaretToAttribute() which also restores the gravity there. I think that it might be better if that util always handled restore manually.

I'm inclined now towards rewriting this logic with these assumptions:

• bindTwoStepCaretToAttribute() has its own flag whether it has just overridden gravity or not,
• bindTwoStepCaretToAttribute() should also take the responsibility to restore the gravity when needed,
• we completely remove the auto-restoring logic from document selection,
• we add some assertion logic for cases which should not happen and see if it helps to find bugs.