Conditional permissions

[martinthomson]

As the use of the tools in this spec are being used for more different features, we're starting to see the gaps.

For instance, requests for media autoplay (#172) and fullscreen (#26) highlight the potential need for conveying additional permissions, specifically engagement. The storage access API implemented variously by Apple and Mozilla also shows how engagement with the origin is a necessary precondition for even showing a permissions prompt, and it might be necessary - even with permission - to have engagement before the associated action can be triggered.

The API currently manages this poorly. If a permission is queried, what should it say if a request might be an outright denial in the absence of an engagement gesture, but a prompt if there is a recent click or key press?

The permissions state really isn't sufficient for this.