Regression: Can no longer decode Paradise_Lost.pdf

[nico]

This file from Google Books: Paradise_Lost.pdf

We used to be able to render it. Now we instead error out with

warning: Internal error while processing PDF file: JBIG2ImageDecoderPlugin: Segment refers to dead segment

The diagnostic is right and I think the file is wrong, segment 15 refers to segment 14, but that doesn't have its "Retained?" bit set:

Segment number: 14
Segment type: 0
Page association size is 32 bits: false
Page retained only by itself and extension segments: false
Retained: false
Referred-to segment count: 0
Segment page association: 1
Segment data length: 53921

Segment number: 15
Segment type: 6
Page association size is 32 bits: false
Page retained only by itself and extension segments: false
Retained: false
Referred-to segment count: 2
Referred-to segment number: 0, retained true
Referred-to segment number: 14, retained false
Segment page association: 1
Segment data length: 3946

But we should display it anyways, of course.

I think the things to do are:

1. Add an enum to JBIG2Loader that controls how strict we are, and set it to less strict by default and request more strict in jbig2-from-json
2. Check if jbig2enc (which Google Books uses) still gets this wrong, and if so file an upstream issue about it