After typechecking a tree, the typer adapts it to the current
mode and expected type. If we are in FUNmode (typechecking the
qualifier of a value- or type-application), and the tree does not
already have a MethodType or PolyType, it is reinterepreted as
`qual.apply`.

In doing so, `insertApply` stabilizes the type of `qual`, e.g.
replacing `Ident(foo).setType(typeOf[Int])` with
`Ident(foo).setType(typeOf[foo.type])`.

However, this does not check for by-name parameters, which cannot
form the basis for a singleton type, as we can see by trying that
directly:

```
scala> def foo(a: => String) { type T = a.type }
<console>:7: error: stable identifier required, but a.type found.
       def foo(a: => String) { type T = a.type }
                                         ^
```

When I last touched this code in SI-6206 / 267650c, I noted:

    // TODO reconcile the overlap between Typers#stablize and TreeGen.stabilize

I didn't get around to that, but Adriaan gave that code a thorough
cleanup in fada1ef.

It appears that on the back of his work, we can now remove the local
stabilization logic in `insertApply` in favour of `TreeGen.stabilize`.
We then avoid the ill-formed singleton type, and the spurious
"apply is not a member" type error.

I did have to modify `isStableIdent` to check the symbol's info
in addition to the tree's type for by-name-ness.