Trim descriptions based on bytes, not characters

It turns out that the Hub description length limit is 25k *bytes*, not characters.  We've gotten lucky that our few unicode glyphs have landed so far such that we've never run into a problem with this before now, but we're currently hitting this while trying to push the `amd64/nextcloud` description in that the trimmed version ends up being exactly 25,003 bytes thanks to the unicode glyphs throwing off our trimming calculations (and getting really [un]lucky with the exact positioning of the markdown such that our trimming doesn't walk back far enough).

This fixes that by calculating and trimming based on bytes, accounting for cases of unicode glyphs directly spanning the 25k byte boundary.

After applying this fix, the same breaking `amd64/nextcloud` trimmed description is 24,725 bytes instead.

push.pl

@@ -15,7 +15,9 @@ use Mojo::Util qw(decode encode trim url_escape);
 use Term::UI;
 use Term::ReadLine;
 
-my $hubLengthLimit = 25_000;
+require bytes; # this is not recommended, but we *only* use "bytes::length" from it to determine whether we need to do a more correct conversion to/from bytes for trimming (see $hubLengthLimit and usages)
+
+my $hubLengthLimit = 25_000; # NOTE: this is *bytes*, not characters 🙃
 my $githubBase = 'https://github.com/docker-library/docs/tree/master'; # TODO point this at the correct "dist-xxx" branch based on "namespace"
 
 my $username;
@@ -75,7 +77,7 @@ sub prompt_for_edit {
 		$proposedText =~ s%$supportedTagsRegex%$sponsoredLinks$1$2%;
 	}
 	
-	if ($lengthLimit > 0 && length($proposedText) > $lengthLimit) {
+	if ($lengthLimit > 0 && bytes::length($proposedText) > $lengthLimit) {
 		# TODO https://github.com/docker/hub-beta-feedback/issues/238
 		my $fullUrl = "$githubBase/$proposedFile";
 		my $shortTags = "-\tSee [\"Supported tags and respective \`Dockerfile\` links\" at $fullUrl]($fullUrl#supported-tags-and-respective-dockerfile-links)\n\n";
@@ -91,9 +93,18 @@ sub prompt_for_edit {
 		$trimmedText =~ s%$supportedTagsRegex%$sponsoredLinks$1$tagsNote%ms;
 		# (we scrape until the next "h1" or a line starting with a link which is likely a build status badge for an architecture-namespace)
 		
-		if (length($trimmedText) > $lengthLimit) {
-			# ... if that doesn't do the trick, then do our older naïve description trimming
-			$trimmedText = $startingNote . substr $proposedText, 0, ($lengthLimit - length($startingNote . $endingNote));
+		if (bytes::length($trimmedText) > $lengthLimit) {
+			# ... if that doesn't do the trick, then do our older naïve description trimming (respecting utf8; see https://www.perlmonks.org/?node_id=1230659 and https://perldoc.perl.org/utf8)
+			$trimmedText = $proposedText;
+			utf8::encode($trimmedText);
+			$trimmedText = $startingNote . substr $trimmedText, 0, ($lengthLimit - bytes::length($startingNote . $endingNote));
+			# (deal with the potential of "bytes::substr" here cutting us in the middle of a unicode glyph, which is arguably a much worse problem than the markdown cutting described below 😬  again, see https://www.perlmonks.org/?node_id=1230659)
+			$trimmedText =~ s/(?:
+				[\xC0-\xDF]
+				| [\xE0-\xEF] [\x80-\xBF]?
+				| [\xF0-\xF7] [\x80-\xBF]{0,2}
+			)\z//x;
+			utf8::decode($trimmedText);
 
 			# adding the "ending note" (https://github.com/docker/hub-feedback/issues/2220) is a bit more complicated as we have to deal with cutting off markdown ~cleanly so it renders correctly
 			# TODO deal with "```foo" appropriately (so we don't drop our note in the middle of a code block) - the Hub's current markdown rendering (2022-04-07) does not auto-close a dangling block like this, so this isn't urgent