Failures on [tests] Use asyncStart/asyncEnd for two language tests about sdk

[mkustermann]

/cc @sigmundch for routing to ddc team

[lrhn]

The code is calling _asyncCompleteWithErrorCallback, which (after a likely no-op check) calls _asyncCompleteError, as the stack trace shows.

The _asyncCompleteError function starts with an assert(!_isComplete);, which the code passes, so the _Future doesn't have a value/error yet.
It may be either pending complete, chained or just incomplete. It should be incomplete, but that's not what's being tested, so that is suspicious by itself.

It then calls _setPendingComplete which tries to set the future to be pending completion (not completed yet, because it's asyncComplete, but can't be completed with something else any more, just waiting impatiently to complete with the already provided error.)
That function starts with assert(_mayComplete);, which fails. That check only accepts being incomplete.

Which means the future state is either chained (has been completed with another future, waiting for that) or pending complete (has been asynchronously completed, haven't gotten around to actually do it yet), at a point where the async code tries to complete the same future again.

It's not completely impossible that this is related to a bugfix I made recently, or at least which asserts are used, because I do call _asyncCompleteError from somewhere that also doesn't check for _mayComplete. I'll give the state transitions a look-over.

(It's an error test, so most of these individual tests should not even compile. The unchecked_use_of_nullable_test_02 one should, though, it's marked as # 02: ok.)

[nshahan]

@lrhn
From an in person discussion, DDC always runs with assertions enabled in the test application. Do any of these tests fail on any other platforms if you run them with assertions enabled? That could help us track down if the failures are related to your change.