I don't think we can do this quite like this in this codepath - this means this fast path will cause us to skip jsdoc type assertions in JS, which look like /* @type {Type} */(expression).

Same example as repro, but in JS:

// @ts-check
/**
* @typedef {{message: {id: string}}} DataShape
*/

/**
* @param {string | number} id
* @return {any}
*/
function getObject(id) {
  return {} as any
}

;(() => {
  /* @type {string | number} */
  let id = 'a'
  while (1) {
    const data = /* @type {DataShape} */(getObject(id))
    const message = data.message
    id = message.id
  }
})()

We can do it this way, it just means that we won't bail out early during control flow analysis. It's simply a matter of how many cases we feel are worthwhile handling in getTypeOfExpression. In a TypeScript file it matters because you'll get an error with --noImplicitAny. In a JavaScript file it just means another any, which are a dime a dozen already, so to speak. So, I think it is fine.

noImplicitAny still works in JS files with checkJs, so the bug does still exist in JS then, it just has a smaller audience.