…using the package-lock.json from beta and re-running npm@8.5.5 install

If any of these functions are called while router is still null, that
means our setTimeout(()=>…) dependency injection didn't work.

Let's make that show up as an error in the browser console, rather than
silently failing to navigate.

I'd like to make sure whenever we update this library, we do it as a
focused and intentional endeavor, where someone gets to read the patch
notes, and decide how much manual QA is needed.

Patch notes for react-router:

- https://github.com/remix-run/react-router/releases

That might be true even if we weren't using `unstable_` API's.

An alternative would be to use `~6.14.1` to stay on `6.14.x`, allowing
some `npm` commands to sometimes update patch versions if they exist.

---

Package-lock.json already specifies 6.14.1 prior to this commit, because
it updated while I was resolving the merge conflict (6.14.1 qualifies as
^6.13.0).

#3012 (comment)

Language selector had an extra box outline introduced by #4207
and the click height was short.

(UI fix not related to current PR, this came up while testing the
form processing route.)

Superceded by /forms/:uid/settings/media

formatting-only commit

- Manually hoist variables to make scope clearer and use 'let'
- Carefully replace 8 loose comparisons with strict ones (===, !==)
  - https://dorey.github.io/JavaScript-Equality-Table/ was useful here

It seems asset.content.translations can be undefined if you have a form
with 0 questions, which is pretty uncommon but it leads to a full-page
crash here.

Fix the same crash when it appears in the data table settings menu,
and on the data reports page.

…as it still has some impact on subroute behavior.

https://github.com/remix-run/react-router/releases/tag/react-router%406.14.2