Type checking and simplification/cleanup of tests

[eyelidlessness]

What

This change does a few things that aren’t totally inseparable, but have a bunch of chicken/egg characteristics that make it feel more appropriate to land them together:

• Add static typing to all tests: this is a bit more abrupt than the gradual story discussed in [issue I neglected to file here, but filed in Core]. The reason I've started with tests is that I'm highly motivated to thoroughly understand the existing tests, in preparation for some work that should follow pretty closely behind.

• Adds some known type definitions¹ for libxmljs, which is simultaneously critical and particularly opaque/risky for both maintenance and more forward-looking efforts.

• Sets up Vite as a build tool for TypeScript, and Vitest as its test runner. I previously converted core and express to ESBuild, which Vite uses for most transpilation duties. It turns out that Vite is simpler to set up (a breath of fresh air in JS ecosystem build tooling), and Vitest being so trivial to integrate with nearly no config fuss is a big testament to that. Vite also has some very compelling features around targeting multiple environments. For today, what that means is that while I deferred converting the user-facing portion of transformer (in this PR; that will follow shortly, but this one is already big enough), it should be pretty straightforward to handle the build and distribution portions of that effort.

• Simplifies test code in a variety of ways. Mostly this is chore work cleaning up copypasta and patterns that make tests harder to follow (e.g. converting Promise API calls to async/await, direct property access rather than an opaque chai extension). A smaller portion, but one I think will be more valuable in coming days, isolates some platform-specific stuff.

Why [now]?

Partly because this accomplishes some of what I’ve wanted to do to be a better maintainer in the long term. If I can better understand the code, I can better address user needs, and find root causes of bugs more quickly. Static typing helps with that immensely, and means I can do future work more confidently with less risk of introducing regressions. If I called it a year on this PR alone, I’d feel happy about wrapping up on that note. But there is more to come!

Footnotes

1. Well, relatively safely inferred from usage. ↩

[eyelidlessness]

It's a holiday already in some parts of the world, so I put off some commit history hygiene which will hopefully make this change as reviewable as possible. But in the spirit of it being a holiday, I left this little gift.

[eyelidlessness]

I'm not sure how much better the diff is going to get at the pull request level, but I've tried to at least make sure that file renames are part of the commit history and that you can at least view diffs for each file by commit, rather than a deleted and added file. If I've missed any and it would hinder review, I'm happy to give it another shot.

The history still isn't necessarily ideal, particularly where bad-external.spec.js was changed to use shared logic while also converting transformer.spec.js. Those changes felt logical at the time but may not be great for review.

So, in general, if it would help review to further revise the commit history I'd be happy to do that. But I'm hesitant to get more bogged down in that without direction from a reviewer.

[lognaturel]

The TODO in 9b1dc9a is no longer relevant, right? It looks like it was easy to keep the badge and if it is I think we might as well keep it.

I went through each commit to make sure I understood the general nature of the changes introduced and the reason for them. I did NOT verify the types myself and did NOT read through every single transformation. I paid special attention to bigger sections of diffs or diffs within a file that didn't match an established pattern. Overall it seems unlikely coverage has been lost or changed and this is looking ready to merge to me.