Fix typing of new

[odersky]

Assume

   class C[X]
   type A[X] = C[X]

The previous rule for new A overwrote the type of A
with C, in order to satisfy the condition that the type
of new should be a class type. Overwriting types like
that in Typer is wrong, since it is not preserved by pickling.
We tried to patch up things in RefinedPrinter, but this is not
enough, as test case alias-new.scala shows. This test caused
the type in a new to be unpickled as a <notype>.

The fix is to treat eta-reduction in new types in a purely functional
way, applied in both Typer and TypeAssigner.

The problem was encountered when systematically testing creation methods,
but it has nothing to do with them.

[odersky]

The fix breaks FromTasty tests. i0306.scala first, and once that it blacklisted a bunch of others where the decompiled source is not compilable. Wrt the example above, we'd see new nodes like:

 new ([X] => C[X])[String]

We should leave this as either new A[String] or new C[String], but expanding A alone to
[X] => C[X] does not work. I am not sure where this is done, though. @nicolasstucki can you take a look?

Note: Previously this problem was hidden because we overwrote the alias type A with C in Typer.

[nicolasstucki]

Now i0306.scala passes the tests