home: the neural ledger leads with Arabic

The front page's model list omitted Arabic entirely - the campaign's
headline result. It now leads with the teacher (ara-diac-2.0, DER 2.29,
best dedicated model on SadeedDiac-25), the released client tier
(ara-diac-small-2.1, int8 264 MiB, DER 4.57), and the browser tier
(ara-diac-layerdrop-1.0, int4 95 MiB, DER 5.78), ahead of the existing
Hebrew, Thai, and Urdu rows. All figures ledger-sourced.

Author: ronaldtse

src/pages/index.astro

@@ -264,6 +264,21 @@ for (const code of systemCodes) {
         </p>
       </header>
       <ul class="neural-ledger">
+        <li class="neural-row">
+          <span class="neural-task">diacritization</span>
+          <code class="neural-id">ara-diac-2.0</code>
+          <span class="neural-metric tnum">DER 2.29 · best dedicated on SadeedDiac-25</span>
+        </li>
+        <li class="neural-row">
+          <span class="neural-task">diacritization · client</span>
+          <code class="neural-id">ara-diac-small-2.1</code>
+          <span class="neural-metric tnum">int8 264 MiB · DER 4.57</span>
+        </li>
+        <li class="neural-row">
+          <span class="neural-task">diacritization · browser</span>
+          <code class="neural-id">ara-diac-layerdrop-1.0</code>
+          <span class="neural-metric tnum">int4 95 MiB · DER 5.78</span>
+        </li>
         <li class="neural-row">
           <span class="neural-task">diacritization</span>
           <code class="neural-id">heb-diac-1.0</code>