Stop Oxc::Node from shadowing the ESTree fields it reads

[marcoroth]

This pull request fixes three fields that Oxc::Node was hiding behind methods of its own, and teaches a field to answer to its snake_case name.

A node reads a field through method_missing, so every method it defines shadows a field called the same thing. Three of them did, and each answered something plausible instead of raising.

Oxc.parse("let count = 0").root.every("Identifier").first.name
#=> "identifier"

Identifier#name is the one that matters, since it is the most common read in any codemod. JSXElement#children answered the opening and closing elements instead of what the element wraps, and ImportDeclaration#attributes answered the wrapper's own hash instead of the import's with clause. Nothing raised in any of the three, so the wrong answer just flowed onward.

What makes a name safe

ESTree spells its fields in camelCase or as a single lowercase word, so a method name carrying an underscore can never collide with one.

node.underscored_type   # was `name`
node.child_nodes        # was `children`
node.to_h               # `attributes` is now protected

Checking a candidate name against @oxc-project/types overreports. It declares parent on all 121 interfaces and the serializer emits it on none of them, so the only reliable check is what a parse actually returns.

Reading a field the Ruby way

Since ESTree is camelCase throughout, reading an AST used to mean writing JavaScript casing in Ruby. A field now answers to either spelling.

node.type_annotation.type_annotation.type_name.name
#=> "Thing"

root.source_type
#=> "module"

declaration.super_class.name
#=> "B"

respond_to? answers for both names. An unknown field still raises NoMethodError, and a method the walker defines still wins, since method_missing never sees it.