Support VitalSource books which are created from PDFs rather than EPUBs.
These books have the same frame structure as EPUBs, but use one HTML
page per PDF page in the inner frame, rather than one HTML page per
chapter as is more common for EPUBs.

VS PDFs do not contain a selectable text layer. Instead they consist of
a high-resolution rendered image of the current page plus data about the
content and location of text characters embedded in a script tag in the
page which sets a `window.innerPageData` variable.

To support this in Hypothesis, we generate a hidden text layer from the
`innerPageData` data and overlay it on top of the rendered image. This
hidden text layer is similar to what PDF.js creates for us in our
standard PDF viewer.