[test] Misc improvements to experimental and browser test runner

[eps1lon]

1. test:karma: More ergonomic approach to running test:karma with specified browsers

   -`yarn test:karma --browsers BrowserStack_Chrome,ChromeHeadlessNoSandBox,BrowserStack_Safari,BrowserStack_Firefox,BrowserStack_Firefox`
   +`yarn test:karma --browsers chrome,chromeHeadless,safari,firefox,edge`

2. t (experimental cli): Fix shorthand for --bail (-b) and --inspect-brk (-d)
   Both used -b.
3. Explicitly throw if createClientRender or createMount are mistakenly called in test hooks
   This is not problematic right now as far as I can tell but any potential before hook called from within createClientRender would be dropped if createClientRender is called within a before hook.
   Only Divider.test violated this.
4. t: Allow testing in production environment.
   Does not work for all tests. Adding it now to gradually works towards enabling tests in prod.
   Tests should not do/assert something that actual users will never see. By only running in development we risk moving away from the "real world".
5. test:karma: Remove outdated aliases
   These are no longer needed with [core] Ship modern bundle #22814
6. test:karma: Fix known process.env.* variables being replaced with an object instead of their hardcoded (string) values
   This currently defeats dead-code-elimination and gets into some weird edge cases if we read from process.env which becomes truthy in browser tests.

[eps1lon]

was afterEach(async () => {) which we no longer need now that cleanup is sync.

[mui-pr-bot]

No bundle size changes

Generated by 🚫 dangerJS against 8948c5c