Better data-driven integration test runner

[ahdinosaur]

Describe the solution you'd like

If the data-driven integration tests fail, we should know exactly which test case caused the failure.

At the moment, one #[test] runs multiple data tests, so it's hard to know what exactly went wrong.

We should improve this somewhat.

Design inspiration

• JSON-e uses a build.rs to build a test file based on their spec data. At first I was resistant, but maybe that is the best approach.

Describe alternatives you've considered

The alternative is we write a re-usable test macro. Myabe that's better.

Additional context