clean up and refactor

[caindy]

The immediate impetus for this change was to use the application/problem+json response for 400s, but I took the opportunity to apply the head/get refactoring for those paths that have HEAD defined.

I was careful not to introduce or to remove a 410 Gone response where it seemed inappropriate. Reference data comes to mind.

There remains a need to clean up or refactor a few paths that were left mostly as-is because they didn't have a HEAD.

[caindy]

I have a much cleaner what to reuse responses that I'm going to push to this branch today.

[caindy]

There are still a couple of constants to pull out, but this is substantively done. I was able to remove 1200 lines and get rid of a lot of warnings from the linters.

[caindy]

The main things here are:

• Organize collection schemas into array-max-*.yaml files. I tried to name the files after the natural upper bound that they all share. This makes it clearer what the collection describes, too, besides not having to repeat the actual constant.
• Move most of the base responses into the top level openapi.yaml file. This means that all the header definitions can be references in the bundled spec which had ballooned past 12k lines. It's now down to ~4500, which is what the original unsplit spec sat at, IIRC.
• Enable redocly lint again locally on save. We can add the VS Code plugin back to populated the problems tab, but this is working for me. That linter validates examples. I deleted a couple of examples that weren't validating, but we have a lot of examples to create, so I don't think that's a problem.
• Ensure all requests use 400 with the problem media type, and they all have an empty 500 which is used as the default response, too.