Remove hidden nodes from the acccessibility tree

[hovinen]

This is as per https://www.w3.org/TR/wai-aria-1.2/#tree_exclusion. It also facilitates testing: in cases where there is only one non-hidden element of a particular role, one can just query for that role without having to specify the query more narrowly.

This includes unit tests of the functionality. To facilitate the assertions it adds a dependency on the crate Test That!.

WPT results

No changes in test results compared to main.

_{Generated by the WPT workflow.}

[nicoburns]

Core change looks good, but some notes.

[nicoburns]

Not sure about the use of the library here. It introduces an extra DSL that needs to be learnt, and it doesn't seem like it'd be much shorter than a pure-rust assertion...

[hovinen]

Of course it's your call. If you want, I will replace it with a built-in assertion.

I found it easier to write the assertions this way. I can write the precise assertion I want in the test without over- or underspecifying, and I get a context-rich failure message for free.

There's an example showing why I care about that in another test which I had to investigate and fix as part of this change: a_semantic_page_has_no_unknown_elements. The assertion looks overspecified compared to the name of the test -- if I take the test name literally, it should be asserting that the elements with unknown Role is a subset of the given set, not a sequence equal to it. But it's not clear whether this is intentional -- is it important that there be a head element with unknown Role in the accessibility tree? Did I break something? (For now, I've assumed that the test name is correct, since that seems to make sense, but I'm really not sure.)