Remove the document-wide image access API

We only use the per-page image access API nowadays. This maps better to
how Poppler does this, and it paves the way for full on-demand parsing
of the PDF content.