The previous pattern to use the same `assert` and `require` objects for
every sub-test in a test led to false output and failure reporting,
because the nested `t` wasn't being used, and therefore the context was
lost. So create `assert` and `require` objects inside every `t.Run()`
function.

It's slightly more descriptive and frees up `name` for uses relevant to
the code being tested.

And upgrade the macOS Intel runner.