Expand defer tests

[phughes-scwx]

As requested in #3384 (link), this updates the defer tests to cover using the if argument, and the label argument for fragment spreads.

Additionally, this PR:

• refactors the defer tests into a table test so that we can do more fine-grained, duplicative tests without repeating handler / response parsing boilerplate;
• updates the defer tests to run every test against both transports that support @defer, not just SSE, to ensure consistency between the implementations;
• adds a test client helper Client.IncrementalHTTP based on Client.SSE that can be used to test@defer against transport.MultipartMixed.

General notes:

• I've added a some documentation glosses that may or may not be accurate / useful. Correct me or update as you see fit.
• I am not totally sure of the multipart/mixed client implementation: did my best on that one. It's in the documentation and TODO notes, but I may have missed the point in some cases.
  • If you want, we can break this part of the PR out and keep the tests focused on SSE for now.
• In order to reduce duplication for the table tests I made a few choices (like using a generic type and reflect) that can be obscure, and you may not like that and it can be replaced. But my personal preference is to make a clear assertion that is encoded in the table test case struct, and this made it possible with a lot less repetition.

@StevenACoffman @giulio-opal

[phughes-scwx]

Not certain why the coverage job is failing: I regenerated the singlefile test server to rename the operations in the test... May be there is some golden test for the server codegen I do not know how to update?

[phughes-scwx]

This Thanksgiving I am grateful for all those who maintain open source projects. 👍

[StevenACoffman]

Hey, thanks for circling back to this! I'm on my phone, but it looks like something is not quite right with the tests?

[phughes-scwx]

Yeah no hurry or anything, but I don't quite know what I'm missing. Getting hundreds of "results did not match" on the schema-generated code for the singlefile test. If I had to guess you have a snapshot/goldenfile test for that code, but I didn't have time to track it down. Will look again today.

[coveralls]

coverage: 59.877% (+0.1%) from 59.732%
when pulling 35b0395 on phughes-scwx:issue-1972-follow-up
into de8ff90 on 99designs:master.

[phughes-scwx]

OK, needed to update the followschema codegen test schema to match. Looks good. Will rebase on master but the coverage also looks good.

[StevenACoffman]

Thanks! Sorry about the tab character. This is extremely high quality. I really appreciated that you added such nice comments, especially since this is a draft spec, and you provide a nice trail of breadcrumbs for people to update it if the spec changes.