fix: throttle markdown lexer during streaming to prevent rendering backlog

- Add leading-edge throttle (150ms) on marked.lexer() during streaming
- Reduces lexer calls from 60/sec to ~7/sec, preventing CPU backlog on mobile
- When done=true, throttle is bypassed for immediate final render
- Loaded/finished chats render identically to upstream

Author: Algorithm5838

src/lib/components/chat/Messages/Markdown.svelte

@@ -34,6 +34,7 @@
 	export let onTaskClick = () => {};
 
 	let tokens = [];
+	let lexThrottle = null;
 
 	const options = {
 		throwOnError: false,
@@ -53,13 +54,16 @@
 		]
 	});
 
-	$: (async () => {
-		if (content) {
+	// Throttle lexer during streaming to prevent rendering backlog
+	$: if (content) {
+		if (done || !lexThrottle) {
+			clearTimeout(lexThrottle);
 			tokens = marked.lexer(
 				replaceTokens(processResponseContent(content), model?.name, $user?.name)
 			);
+			lexThrottle = done ? null : setTimeout(() => { lexThrottle = null; }, 150);
 		}
-	})();
+	}
 </script>
 
 {#key id}