Initial implementation of constructor tearoff support

[jcollins-g]

Beginning of work for #2655. This is just enough so that if someone switches on constructor tearoffs, dartdoc is not likely to instantly let loose the magic smoke. So it does not support any of the new syntax and still requires users to use the old syntax to refer to default/unnamed constructors.

[srawlins]

Hmm, why is that? Unnamed constructors and "default" constructors are different things (the latter being a constructor which is implicitly created when a class features no constructors).

[jcollins-g]

Mostly because it is confusing and the two ideas are used interchangeably here. I'll try and clarify.

[srawlins]

Ah interesting. Yeah we'd have to see if it is important to distinguish. For example:

class C {
  C(int i);
}

has a constructor whose isUnnamedConstructor is true, but it is not a default constructor (it cannot be implicitly called by any subclass's constructor).

[srawlins]

Typo, I think: [D.new]

But this is interesting. It is illegal to tearoff D.new, since it is abstract (it cannot be directly called). But folk may still want to refer to it. Because of this, maybe [D.new] should be left as a PrefixedIdentifier, rather than a ConstructorReference, in the analyzer's AST...

[jcollins-g]

Both are intended to work, I'll add the other.

As for the AST implications, I'm guessing it might be a good idea there, yes, for comment reference resolution as well as possibly dealing with erroneous code more cleanly? I don't know the extent to which this is important for those cases.

[jcollins-g]

OK, this is now legit ready for review. It includes migrating the older experiments into the mainline test suite so that we can have experiments that only work in 2.15 without breaking our beta/stable testing.