Gracefully handle invalid UTF-8 sequences in Markdown links

Closes #6077

CHANGELOG

@@ -13,6 +13,7 @@ v 8.4.0 (unreleased)
   - Autocomplete data is now always loaded, instead of when focusing a comment text area
   - Improved performance of finding issues for an entire group
   - Added custom application performance measuring system powered by InfluxDB
+  - Gracefully handle invalid UTF-8 sequences in Markdown links (Stan Hu)
   - Bump fog to 1.36.0 (Stan Hu)
   - Add user's last used IP addresses to admin page (Stan Hu)
   - Add housekeeping function to project settings page

lib/banzai/filter/reference_filter.rb

@@ -133,6 +133,7 @@ def replace_link_nodes_with_text(pattern)
           next unless link && text
 
           link = CGI.unescape(link)
+          next unless link.force_encoding('UTF-8').valid_encoding?
           # Ignore ending punctionation like periods or commas
           next unless link == text && text =~ /\A#{pattern}/
 
@@ -170,6 +171,7 @@ def replace_link_nodes_with_href(pattern)
 
           next unless link && text
           link = CGI.unescape(link)
+          next unless link.force_encoding('UTF-8').valid_encoding?
           next unless link && link =~ /\A#{pattern}\z/
 
           html = yield link, text