Fix and change a11y tests

[gabalafou]

Some of the passing accessibility tests broke because of the way I had implemented the data regression fixtures for pytest-regressions in #1501.

Specifically, some of the data regression fixture files contained the Axe-core version, so when Axe-core was updated, those fixtures became out of sync with the test run.

I decided to scrap the data regression fixture files and instead, I define a function in the test file that filters out violations for Axe-core rules that we want to ignore. Using a function is ultimately more flexible.

[gabalafou]

cc @trallard

[trallard]

Added only a minor comment, but this LGMT plus this approach seems a lot more maintainable than the previous one so this is a +1

Will go ahead and approve and wait a day or so before merging to allow for any other comments

[drammock]

this looks like a nice improvement, thanks @gabalafou!

Long-term, I might consider splitting out the mapping of pagename -> expected/tolerated violation into a dict or other data structure that is external to the function, or maybe even storing it as a separate JSON or YAML file. I usually find that easier to maintain than the current design, where the various exceptional url_pathnames are written into the function code body.

That said, the landmark-unique situation is complicated and might not be easy to abstract out in the way I've just described. So I think for now let's go with what you've got here.