Streams: update tests for Web IDL conversion

[domenic]

Follows whatwg/streams#1035. Notable changes:

• Updates for the normative changes listed there.
• Introduce an idlharness test
• Remove various tests for things that are covered by idlharness, such as brand checks, the order of getting values from dictionaries, etc.
• Updated for the fact that everything is now globally exposed, so some roundabout code to get constructors can be removed.
• Slight timing updates: the pattern of returning a promise from an underlying sink/source/transformer start() function, then waiting on that before doing asserts, does not work with Web IDL's "promise resolved with". So instead we use flushAsyncEvents() to wait a little longer.
• Consolidated queuing strategy tests into a single file, since after deleting the things covered by idlharness, most of the tests are shared.
• Added tests that underlyingSink/underlyingSource are converted after queuingStrategy, since those dictionary conversions are done in prose.
• Added a test for various updates to the Web IDL async iterators specification

[ricea]

lgtm

The removed tests will be done by idlharness.js, so we definitely don't want to keep them.

As you say, it would be good to verify that the second argument is unpacked before the second, but we don't need all this infrastructure to do it, just something like

assert_throws_exact(error2, 
    () => new ReadableStream({ get start() { throw error1; } }, { get size() { throw error2; } }));

should do it.

[ricea]

Everything still lgtm. I don't know what to do about ReadableStreamBYOBRequest. Could we just call idl_test() asynchronously after grabbing one?

[LukeZielinski]

Looks like TaskCluster didn't run here, retrying.

[stephenmcgruer]

Looks like TaskCluster didn't run here, retrying.

This is a draft PR, TaskCluster doesn't run on drafts.

[ricea]

I re-reviewed everything. Still lgtm.

[ricea]

Updates lgtm.

[TimothyGu]

The recent additions around next/return interactions look good, modulo one nit. Also, would it be a good idea to have an assert_iteration_result(), perhaps for future?