I won't try to do anything special here yet, instead just calling
it imperatively. Sort this out later.

This is premature optimization and makes things way more difficult
to reason about.

- anchor / describe both take options
- TextQuoteAnchor takes position as an option but it's not stored
- helper closures sprinkled throughout have fewer _ prefixes

It'll cause more false positives than it helps.

Also change the PDF anchoring to leverage the HTML anchoring by
allowing the options to be passed into the module-level anchor
functions in the pdf and html modules so that the pdf module can
be only responsible for finding the appropriate page and then
delegating.

Avoids falsely anchoring to positions that won't resolve once a PDF
page renders.

In order to clean up vocabulary and better serve the needs of dynamic
pages, refactor `setupAnnotation` to deal with re-setup, unhighlighting
targets that have been removed from the annotation and re-anchoring
anchors that have lost their ranges.

- Store the Range on the anchor

  * The BucketBar would use it for getBoundingClientRect, except that
    it's a bit broken on Chrome.

  * The PDF Plugin can delete it to trigger re-anchoring no page load

- Use MutationObservers in the PDF plugin and make smarter judgements
  about when pages are done rendering by using the renderState in
  addition to the text layer renderingDone flag.

Sometimes the context might hinder the anchoring if the context has
changed but the quote has not.

TextPositionAnchor can be generated and re-anchored within a PDF so
long as the selection doesn't span a page boundary.

TextQuoteAnchor is generated, but not re-anchored.