Remove tink_streams dependency.

[back2dos]

I forgot why exactly we decided to model everything with streams - other than the fact that it's super fancy. If there's a good reason to, then we should of course keep things as they are. Otherwise I'm inclined to remove them for the sake of simplicity.

A non-breaking approach would be to back Case by CaseImpl (better name wouldn't hurt), which looks like CaseObject but has a different execute method:

• function execute(ctx:{ function report(a:Assertion):Void; function finish():Void; )->Void
• function execute((report:Assertion)->Void)->Future<Noise>

I'm leaning towards the former one, because it's simpler still.

An implicit cast @:from CaseObject would then create an adapter that unrolls the returned Assertsions into the reporting function.

If we're fine with a breaking change, we can make the signature change on CaseObject directly. Since the main usage of this lib is via tink_unittest, it should be possible to leave all test suites based on it unaffected.

[kevinresol]

I think I originally started from a mind model that Case = Array<Assertion>, Suite = Array<Case>, etc (and Array could be async thus become Stream). That pure/functional approach was easier for me to reason about at a top-level view. But I think I do agree that makes the internal implementation more complex.

Anyway I think no one really use this lib directly so I guess a breaking change is fine. (if anyone actually does he will be a hardcore anyway and I guess wouldn't hurt much from a breaking change)