Update openhtmltopdf. Add visual regression test script

[jamesbarnett91]

Update openhtmltopdf. We were on quite an old version so there is risk of regression issues.
However, since we now have an API, I've written a quick script to walk the API and generate all labels, and ran it against both this updated branch and the existing QA deployment (which still has the old version of openhtmltopdf). Then ran a pixel based diff of both images to check for any visual regression between the labels generated with the old and new versions of openhtmltopdf.

I've attached the diff images as a zip diff-images.zip (or you can run the script yourself). The image is a composite of diff, old version then new version, left to right. Only labels with any change have been included, so any not in the zip have not changed.

There are some interesting differences but they are all very minor, mostly related to text anti-aliasing. In some cases the new version seems more 'correct', such as the pluses on Range Hoods, or the radiator icon on Solid Fuel Boilers.

As part of approving this PR could you check the image diffs and confirm you're happy?

[matteason]

I'm happy with the diffs that only affect anti-aliasing and fix the radiators

The only real regression I can see is the font on the kg icon on tumble dryers has changed, I think from Verdana to Arial

Before:

After:

But looking at the template and legislation I think it actually should have been Myriad all along anyway:

I didn't spot this before because it's so close to Verdana, but worth fixing now we've seen it. I'll send you the files for Myriad