fix: Remove array support from inferType.

[stephenh]

I believe this was always broken, because the oid of a boolean[] is not the same as the oid of a boolean, and AFAICT inferType was resolving "boolean[] --> the oid of a boolean" without any attempt to translate over to the boolean[] oid.

There's talk of removing inferType entirely, which would have to wait until v4, but I think if this array-based support has always been broken (afaiu?), we can just remove it immediately in a v3 bug fix release.

Fixes #471

[stephenh]

Disclaimer I haven't run the tests, b/c am waiting for the workflow to be approved (i.e. I haven't looked into getting the tests to run locally--would be great to have a CONTRIBUTING.md tldr).