Change how new operators are compiled.

Instead of an instruction that always instantiates an instance,
the "new" is compiled to a method call on the class, followed by a
method call on the (presumably) new instance.

The former lets built-in types like Fn do special stuff instead of
allocating an instance. In particular, it means:

  new Fn { ... }

can just return the block argument.

test/function/type.wren

@@ -1,4 +1,4 @@
-IO.print((fn 0) is Function)      // expect: true
-IO.print((fn 0) is Object)        // expect: true
-IO.print((fn 0) is String)        // expect: false
-IO.print((fn 0).type == Function) // expect: true
+IO.print((fn 0) is Fn)      // expect: true
+IO.print((fn 0) is Object)  // expect: true
+IO.print((fn 0) is String)  // expect: false
+IO.print((fn 0).type == Fn) // expect: true

test/is/is.wren

@@ -1,13 +1,13 @@
 IO.print(Num is Class) // expect: true
 IO.print(true is Bool) // expect: true
-IO.print((fn 1) is Function) // expect: true
+IO.print((fn 1) is Fn) // expect: true
 IO.print(123 is Num) // expect: true
 IO.print(null is Null) // expect: true
 IO.print("s" is String) // expect: true
 
 IO.print(Num is Bool) // expect: false
 IO.print(null is Class) // expect: false
-IO.print(true is Function) // expect: false
+IO.print(true is Fn) // expect: false
 IO.print((fn 1) is Num) // expect: false
 IO.print("s" is Null) // expect: false
 IO.print(123 is String) // expect: false