Provide better parentPath typings

[JLHwung]

Q	A
License	MIT

In this PR we provide better NodePath#parentPath and NodePath#parent typings using a map from node types to its parent AST node types, generated from our AST definitions. There are no runtime overhead shipped to our end users because they are done in typings.

The new typings catches two babel types error because we used the parentPath info in certain plugins but forget to add babel types support.

This PR includes commits from #14737. I will rebase once that PR is merged.

[JLHwung]

Caveat: the Placeholder has a unique role here.

[babel-bot]

Build successful! You can test your changes in the REPL here: https://babeljs.io/repl/build/52470/

[JLHwung]

isReferencedIdentifier ensures that this is either an Identifier or a JSXIdentifier, but ReferencedIdentifier can't handle JSXIdentifier.

[nicolo-ribaudo]

TS already know that initializerPath is a t.Expression, and t.Expression does not include t.JSXIdentifier.

Could we type isReferencedIdentifier like this, so that TS can automatically infer with initializerPath.isReferencedIdentifier() that initializerPath is a NodePath<t.Identifier>?

  isReferencedIdentifier<T extends t.Node>(
    this: NodePath<T>,
    opts?: object,
  ): this is NodePath<T & VirtualTypeAliases["ReferencedIdentifier"]>;

[JLHwung]

In this way the isReferencedIdentifier query can not differentiate a NodePath union, i.e. NodePath<A> | NodePath<B> inferred from path.isA() || path.isB(), see also the typing errors in recent CI. I tried typings like

isReferencedIdentifier<T extends t.Node, T2 extends t.Node>(
    this: NodePath<T> | NodePath<T2>,
    opts?: object,
  ): this is NodePath<(T | T2) & VirtualTypeAliases["ReferencedIdentifier"]>;

but TS can't pick it up. Help is welcome. I think the intersection behaviour is more useful, so from now on I will ignore such typing errors.

[nicolo-ribaudo]

I wonder why our NodePath has the wrong variance and NodePath<A | B> is not inferred as the same as NodePath<A> | NodePath<B> 🤔

[JLHwung]

This is incorrect. A Super's parent must be either a CallExpression or a MemberExpression. We should consider remove Super from Expression in Babel 8.

I didn't notice CI is failing