When a new content frame was found in VitalSource the client was only injected
either if the frame is already loaded, or when the frame next emits a `load`
event. In the latter case this waits until the document and its subresources
have fully loaded. This can be slow in EPUB chapters that have a lot of images.
Improve this by replacing the frame `load` event observer with a call to a new
`onDocumentReady` function which fires as soon as a document becomes interactive
(according to its `readyState`).

 - Rename existing `onDocumentReady` utility to `onNextDocumentReady` to
   make it clear that it only fires once, and change the implementation to
   be a wrapper around a new `onDocumentReady` function.

 - Add new `onDocumentReady` utility which monitors a frame for changes in the
   content document and invokes a callback each time a document becomes _ready_
   (`readyState` is `interactive` or `complete`).

 - Redo the tests for utilities in `frame-observer.js` so that they use
   real iframes rather than fake ones. Iframes have a complex interface
   and loading sequence, so we really need to use the real thing for
   tests to give us confidence.

 - Use `onDocumentReady` in the VitalSource integration to respond more quickly
   to book content loading in a new content frame.

 - Modify several tests for `FrameObserver`, `VitalSourceInjector` and
   `HypothesisInjector` to be less sensitive to exact timings of events,
   as these changed between the previous and new methods for detecting
   when a document is ready.

Fixes https://github.com/hypothesis/client/issues/4270