feat: throttle markdown lexing during streaming

Add leading-edge throttle (50ms) to marked.lexer() calls during
streaming to reduce CPU work on long responses and low-end devices.
When done, renders immediately without delay.

Author: Algorithm5838

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

@@ -1,4 +1,5 @@
 <script>
+	import { onDestroy } from 'svelte';
 	import { marked } from 'marked';
 	import { replaceTokens, processResponseContent } from '$lib/utils';
 	import { user } from '$lib/stores';
@@ -35,6 +36,10 @@
 
 	let tokens = [];
 
+	// Object ref to avoid Svelte reactivity re-triggering on timer mutation
+	const throttle = { timer: null };
+	const THROTTLE_MS = 50;
+
 	const options = {
 		throwOnError: false,
 		breaks: true
@@ -53,13 +58,31 @@
 		]
 	});
 
-	$: (async () => {
+	const lexContent = () => {
 		if (content) {
 			tokens = marked.lexer(
 				replaceTokens(processResponseContent(content), model?.name, $user?.name)
 			);
 		}
-	})();
+	};
+
+	$: if (content) {
+		if (done) {
+			clearTimeout(throttle.timer);
+			throttle.timer = null;
+			lexContent();
+		} else if (!throttle.timer) {
+			lexContent();
+			throttle.timer = setTimeout(() => {
+				throttle.timer = null;
+				lexContent();
+			}, THROTTLE_MS);
+		}
+	}
+
+	onDestroy(() => {
+		clearTimeout(throttle.timer);
+	});
 </script>
 
 {#key id}