feat: render custom channel emoji in descriptions and comments

Add attachmentRuns handling to parse_description so custom channel
emoji (image shortcodes such as :yt:) render as <img> tags instead of
falling back to their textual form.

The commentViewModel path (comments/youtube.cr) and the video description
path (parser.cr) both route through parse_description, which previously
only handled commandRuns (links). This change merges commandRuns and
attachmentRuns sorted by startIndex and emits an <img class=channel-emoji>
for image-type attachments, mirroring the behaviour of the older
content_to_comment_html implementation.

Fixes #5888

src/invidious/videos/description.cr

@@ -40,6 +40,8 @@ private def copy_string(str : String::Builder, iter : Iterator, count : Int) : I
   return copied
 end
 
+private record Run, start_index : Int32, length : Int32, kind : Symbol, data : JSON::Any
+
 def parse_description(desc, video_id : String) : String?
   return "" if desc.nil?
 
@@ -47,49 +49,121 @@ def parse_description(desc, video_id : String) : String?
   return "" if content.empty?
 
   commands = desc["commandRuns"]?.try &.as_a
-  if commands.nil?
-    # Slightly faster than HTML.escape, as we're only doing one pass on
-    # the string instead of five for the standard library
+  attachments = desc["attachmentRuns"]?.try &.as_a
+
+  # If no runs at all, fall back to simple escape
+  if commands.nil? && attachments.nil?
     return String.build do |str|
       content_size = content.ascii_only? ? content.size : utf16_length(content)
       copy_string(str, content.each_codepoint, content_size)
     end
   end
 
-  # Not everything is stored in UTF-8 on youtube's side. The SMP codepoints
-  # (0x10000 and above) are encoded as UTF-16 surrogate pairs, which are
-  # automatically decoded by the JSON parser. It means that we need to count
-  # copied byte in a special manner, preventing the use of regular string copy.
-  iter = content.each_codepoint
+  # Merge commandRuns and attachmentRuns sorted by startIndex
+  runs = [] of Run
 
+  if commands
+    commands.each do |cmd|
+      runs << Run.new(
+        start_index: cmd["startIndex"].as_i,
+        length: cmd["length"].as_i,
+        kind: :command,
+        data: cmd
+      )
+    end
+  end
+
+  if attachments
+    attachments.each do |att|
+      runs << Run.new(
+        start_index: att["startIndex"].as_i,
+        length: att["length"].as_i,
+        kind: :attachment,
+        data: att
+      )
+    end
+  end
+
+  runs.sort_by!(&.start_index)
+
+  iter = content.each_codepoint
   index = 0
 
   return String.build do |str|
-    commands.each do |command|
-      cmd_start = command["startIndex"].as_i
-      cmd_length = command["length"].as_i
+    runs.each do |run|
+      cmd_start = run.start_index
+      cmd_length = run.length
 
-      # Copy the text chunk between this command and the previous if needed.
+      # Copy the text chunk between this run and the previous if needed.
       length = cmd_start - index
       index += copy_string(str, iter, length)
 
-      # We need to copy the command's text using the iterator
+      # We need to copy the run's text using the iterator
       # and the special function defined above.
       cmd_content = String.build(cmd_length) do |str2|
         copy_string(str2, iter, cmd_length)
       end
 
-      link = cmd_content
-      if on_tap = command.dig?("onTap", "innertubeCommand")
-        link = parse_link_endpoint(on_tap, cmd_content, video_id)
+      if run.kind == :command
+        # Handle link commands
+        link = cmd_content
+        if on_tap = run.data.dig?("onTap", "innertubeCommand")
+          link = parse_link_endpoint(on_tap, cmd_content, video_id)
+        end
+        str << link
+      elsif run.kind == :attachment
+        # Handle attachment runs (emoji images)
+        element = run.data["element"]?
+
+        if element.is_a?(JSON::Any) && element["type"]?.try &.as_s == "imageType"
+          image = element["image"]?
+
+          if image.is_a?(JSON::Any)
+            sources = image["sources"]?.try &.as_a
+
+            if sources && sources[0]?
+              source = sources[0]
+              # Get the emoji image URL
+              emoji_url = source["url"]?.try &.as_s
+              # Fallback for clientResource format (badges)
+              if emoji_url.nil? || emoji_url.empty?
+                emoji_url = source.dig?("clientResource", "imageName").try &.as_s
+              end
+
+              if emoji_url && !emoji_url.empty?
+                # Get accessibility label
+                alt_text = image.dig?("accessibility", "accessibilityData", "label").try &.as_s || cmd_content
+
+                # Build the img tag
+                if emoji_url.starts_with?("http://") || emoji_url.starts_with?("https://")
+                  str << %(<img alt=") << alt_text << "\" "
+                  str << %(src=") << emoji_url << "\" "
+                  str << %(title=") << alt_text << "\" "
+                  str << %(class="channel-emoji" />)
+                else
+                  # For internal resource names (badges), skip rendering
+                  str << cmd_content
+                end
+              else
+                str << cmd_content
+              end
+            else
+              str << cmd_content
+            end
+          else
+            str << cmd_content
+          end
+        else
+          str << cmd_content
+        end
       end
-      str << link
+
       index += cmd_length
     end
 
-    # Copy the end of the string (past the last command).
+    # Copy the end of the string (past the last run).
     content_size = content.ascii_only? ? content.size : utf16_length(content)
     remaining_length = content_size - index
     copy_string(str, iter, remaining_length) if remaining_length > 0
   end
-end
+end