Allow precedence annotations in macro defs

[djrenren]

This pull requests allows for use of precedence annotations inside macro definitions:

    grammar;
    Nothing: u32 = {}
    Expr<B>: u32 = {
       #[precedence(level="0")]
        B,
       #[precedence(level="1")]
       #[assoc(side="left")]
       <left:Expr<B>> "*" <right:Expr<B>> => 0,
    }

    FullExpr = Expr<Nothing>;

This change has two essential parts:

• Extending precedence substitution to include macro names so that <right:Expr<B>> becomes <right:Expr0<B>>
• Forwarding macro parameters when generating the preceding alternative so that we get:

  Expr<B>: u32 = {
    <left: Expr<B>> "*" <right: Expr0<B>> => 0,
    Expr0<B> // <-- Note that B is forwarded to the next level of precedence
  }
  

It seems to work well, my only concern is how to handle substitution of macro arguments in the presence of associativity.
That is, what should we get when have:

Expr<B>: u32 {
   #[precedence(level="0")]
   B,
   #[precedence(level="1")]
   #[assoc(side="left")]
    <left:Expr<B>> "*" <right:Expr<Expr<B>>> => 0,
}

Should it be <right: Expr0<Expr<B>> or <right: Expr0<Expr0<B>>. In this PR we get the former.

My suspicion is that substitution within macro args was already a bit wonky because associative replacement was done in a depth-first manner, skipping the first substitution. This meant that even prior to this change we'd have weird behavior like:

Foo: u32 {
   #[precedence(level="0")]
   "x" => 0,
   #[precedence(level="1")]
   #[assoc(side="left")]
   Bar<Foo, Foo> Foo => 0

becomes:

Foo: u32 {
   Bar<Foo, Foo0> Foo0 => 0
   Foo0
}

So the situation was already messy, but now it's a little worse that the macro name is also a possible substitution. Curious what people's thoughts are.