make l10n tests more reliable

[Thorin-Oakenpants]

some tests require resources which can be too slow to load, resulting in a health failure. Excluding timing, we run region code last to allow as much time as possible. Personally, I never see these, but others reliably do. This will become problematic with automated testing

cc @PieroV

  "l10n_media_messages",
  "l10n_parsererror_direction",  // fine - hardcoded top level document element
  "l10n_validation_messages",    // fine - uses DOMParser
  "l10n_xml_messages",           // fine - uses DOMParser
  "l10n_xml_prettyprint",
  "l10n_xslt_messages",
  "l10n_xslt_sort",              // fine - uses DOMParser

[Thorin-Oakenpants]

html head contains this

	<link rel="preload" href="xml/xmlunstyled.xml" as="script">
	<link rel="preload" href="xml/xslterror.xml" as="script">
	<link rel="preload" href="images/InvalidImage.png" as="image">
	<link rel="preload" href="images/ScaledImage.png" as="image">

elements are hardcoded (offscreen)

		<iframe id="tzpInvalidImage" width="100" height="30"></iframe>
		<iframe id="tzpScaledImage" width="100" height="30"></iframe>
		<iframe id="tzpXMLunstyled" width="40" height="30"></iframe>
		<iframe id="tzpXSLT" width="40" height="30"></iframe>

as soon as the html document has loaded (or rather once we have received the first expected JS file not linked in the html head) we run this code

		// set src's for our l10n iframe tests
			// setting these inline can cause the wrong contentDocument in the wrong iframes
			// it's almost random like some sort of race with different results in android vs windows - WTF!!
			// Switching the element order in html (with inline src and not by JS) I can replicate these
			// mismatched contents in both windows and android. Only setting them via JS are we always correct
			// jesus says: WT actual F
		if (isGecko) {
			try {dom.tzpInvalidImage.src = 'images/InvalidImage.png'} catch {}
			try {dom.tzpScaledImage.src = 'images/ScaledImage.png'} catch {}
			try {dom.tzpXMLunstyled.src = 'xml/xmlunstyled.xml'} catch {}
			try {dom.tzpXSLT.src='xml/xslterror.xml'} catch {} // in FF134 or lower this breaks devtools: oh dear, what a shame
		}

FYI: the images as small as possible (I dare you to make them smaller)

   InvalidImage.png  0 bytes
   ScaledImage.png  81 bytes

once all JS files have arrived, we run the full FP test as such

			let order = [
				3, // first: sets isMB (legacy method)
				2, 1, 5, 14, 13, // fast
				'canvas',
				'storage', // little slow: cache + permissions
				'misc', // cold on load: iframe props
				'elements', // cold on load: mathml
				'audio',
				'webgl',
				'devices','fonts','region', // next to last: allow time for isDevices, font fallback, iframes
				17 // last: uses data collected during gRun
			]

region section is last even after the perf heavy font section

All functions are in region.js

	get_l10n_xml_prettyprint('l10n_xml_prettyprint', isLies),
	get_l10n_xslt_messages('l10n_xslt_messages'),
	get_l10n_media_messages('l10n_media_messages'),

[Thorin-Oakenpants]

none of the three functions have a timeout or attempts to retry - which is what I think we would need to do - timeout and keep trying until we get a valid value (e.g pretty print to be an integer > 20 or whatever, media results to not error, etc )