Improve oxc_traverse

[overlookmotel]

Various improvements to make to oxc_traverse:

• Implement ScopeFlags (ctx.scope()).
• Reduce the size of ancestor.rs by using generics instead of repetition for *Without* types.
• Add TraverseCtx::has_ancestor method where closure returns true or false
  (or maybe Some(true), Some(false) or None)
• Check that TraverseCtx::find_ancestor can return an ancestor node that it found.
• Set enum discriminants for Ancestor so that is_* can use bitwise ops + retag_stack can write 1 byte only.
  • Compiler does not auto-optimize matches!() in is_* functions. Need to do it manually.
  • See ancestor_type2 branch (on my local machine).
  • Need to handle big endian systems where bytes are in reverse order.
    • That's easy, but how to run tests on big endian? Miri?
• Implement Debug for Ancestor and *Without* types
  • Implement intermediate as_ref method which creates a struct of references, which Debug can use.
• Improve API for Ancestor::is_via_*
• API to read siblings in a Vec.
• API to get which index current node is in a Vec.
• API to allow mutating other branches of AST
  • I think all that's required is to:
    • Pass &mut TraverseCtx to enter_* and exit_*
    • Add parent_mut, ancestor_mut methods to TraverseCtx
    • Add span_mut, directives_mut etc to all *Without* types
  • Mutable borrow on TraverseCtx and Ancestor prevents creating more than 1 mut ref at a time
  • Mutable borrow on TraverseCtx unfortunately also blocks calling ctx.alloc().
    • Can solve that with e.g. ctx.ancestry.parent_mut() + ctx.ast.alloc() - separate properties
      can be mut borrowed at same time.
• Move Span to constant position in all structs, for fast lookup?