Replace the usage of `document.title` as a way to get the document title
if the PDF has no embedded title in either its _document info
dictionary_ or _metadata stream_.

In top-level frames using `document.title` (where `document` is the
global HTML document, not the PDF) works because PDF.js sets the title
based on the first non-empty value from:

 1. The embedded title
 2. The filename from the `Content-Disposition` header
 3. The last segment of the URL's path (eg. "test.pdf" in
    "https://example.com/test.pdf")

When PDF.js is embedded in an iframe however, it does not set
`document.title` by default. As a result, documents were ending up in
Hypothesis with a generic "PDF.js viewer" title.

This commit implements (roughly) the same logic that PDF.js uses to
determine the value used to set `document.title`, in the case where the
PDF has no embedded title. This means implementing steps (2) and (3)
from the above list. The `Content-Disposition` filename is not exposed
as a public property on `PDFViewerApplication`, so
`PDFMetadata#getMetadata` was refactored to call the
`pdfDocument.getMetadata` instead.

Fixes https://github.com/hypothesis/client/issues/3372