dispatch:<.=> is busted when used with \foo variables

[zoffixznet]

m: my \foo .= Str
rakudo-moar 72948e847: OUTPUT: «===SORRY!===␤Method call node requires at least one child␤»

# Wrong result; should be 42:
m: my Int \foo .= new: 42; say foo
rakudo-moar 72948e847: OUTPUT: «0␤» 

[jnthn]

I agree it'd be very nice to make this work, though will note that = with sigilless variables is really binding. So we'll have to compile it in such a way that it doesn't use dispatch:<.=>, since that relies on being able to assign. (Alternatively, since for various of the dispatch:<...> methods we've been switching to towards QAST generation anyway, we could just go that way).