When anchoring a quote in a PDF, the quote is first searched in text
extracted using PDF.js's `PDFPage.getTextContent` API, and the resulting
positions are used to create a range within the hidden text layer of a
page.

An issue we've seen several times when doing PDF.js upgrades is minor
changes to which spaces are included in the text layer. In the past
we've adapted our text extraction to match the text layer each time.
This slows down the process of upgrading PDF.js and makes maintaining
compatibility with a range of PDF.js releases more difficult. In the
most recent update, an `includeMarkedContent` option was added to the
`getTextContent` API, and the presence of that option could affect
whether certain whitespaces are included in the output nor not [1].

Try to address this issue generally by mapping offsets from the page
text into offsets in the text layer in a way that ignores whitespace
differences.

 - Add `translateOffsets` utility, which maps a (start, end) pair
   of offsets in an input string into corresponding offsets in an output
   string, where the output is a version of the input that has been
   "corrupted" by the addition or removal of certain characters (eg.
   whitespace)

 - Use `translateOffsets` utility in PDF anchoring to map quote offsets
   in the page text returned by `PDFPage.getTextContent` into offsets in
   the `textContent` of the text layer element.

[1] https://github.com/hypothesis/browser-extension/pull/799#issuecomment-1079864595