Make the actor/object construct into a dec

[rossberg]

This allows simply writing

actor A {
  ...
};
object O {
  ...
};

as a declaration (binding A, 'O', resp.), which avoids the extra let and may make the code look a bit more "natural" to demo.

Currently it is based on a small hack in the parser. I spent the better part of the day trying to implement a more general refactoring of recursive binders (which would desugar them all into blocks, including func and class, turning them into expression forms without their own ids), but that turned out to be harder change than I thought. So I'm deferring that til after the holidays. :)

Edit: for now new and object are simply synonym keywords. Need to think about what to do with them after this.

[crusso]

Ditch new.

[rossberg]

Ditching new sounds plausible, but I'd want to do that as a separate step, possibly introducing some more lightweight sugar for inline object literals.

[crusso]

I'm happy to merge, even if we change this later, it'll be useful for next week.

[paulyoung]

Should actor Foo {...} and object Foo {...} also define a type Foo? (like class would)

[rossberg]

@paulyoung, interesting question. However, objects define singleton values. When would you need to abstract over a singleton value, so that you'd need to mention its type?

[nomeata]

For example with

object Flags {
  var verbose = false;
  var …
}

This is a singleton, yet it would be really nice to use conf : Flags in my function signatures.

[rossberg]

@nomeata, why stop there? Why not do the same for functions?

Let's not go overboard here into the obscure. We should keep type and term definitions separate. The only reason class does both is that that's what it traditionally does. If we remove nominal typing, btw, then it effectively becomes its only purpose: being sugar for combining both a type and a function definition. But it's intentionally an explicit separate form and not built into functions themselves.

[crusso]

I'm happy for you to merge Andreas