tests: look for regressions when converting PDFs

[almet]

I convert all the documents we have in our test suite and store them in a reference folder, and then compare this bit to bit, using pymupdf pixel buffers.

We could use another diffing tool and tell it what's acceptable (if that exists), but I believe that in the end what matters most is the developer experience when we have an output that changes.

Two things come to mind:

1. Inspect the differences
2. Updating the reference version(s)

Inspecting the diff

There are multiple tools that allow to do that, but I found that diff-pdf good and able to generate an output we can look at without having to run a GUI.

diff-pdf /tmp/pytest-of-alexis/pytest-current/sample-docx0.pdf ./tests/test_docs/reference/sample-docx.pdf -m --output-diff=diff.pdf

Produces this diff.pdf file for the changes between the 0.8.1 release and this commit.

Update the reference version

We should have a command to bump all the reference documents (or a specific one).

---

Status:

This PR currently only fails tests when there a change in the output. I plan to do the following:

• Check that PDF outputs are the same (pixel comparison) in our tests
• Collect all differences and publish them as an artifact so we can inspect them, probably as part of the CI.
• Add a tool to update all the reference documents.

Fixes #321