Fix cost section runtime inconsistency and feature card label

- Updated cost section from 4.2s to 5.3s for 2M-node runtime
  (matching the updated 256-dim benchmark data for roadNet)
- Changed feature card heading from "Built-in Classifiers" to
  "MLP Classifier" for specificity
- Scale factor already corrected: 0.111s to 5.3s (~48x)

Author: JacekDabrowski1

website/templates/index.html

@@ -467,7 +467,7 @@ <h3>7 Alternative Algorithms</h3>
             <div class="feature-icon-svg">
                 <svg width="24" height="24" viewBox="0 0 24 24" fill="none" stroke="var(--accent-bright)" stroke-width="2"><circle cx="18" cy="5" r="3"/><circle cx="6" cy="12" r="3"/><circle cx="18" cy="19" r="3"/><line x1="8.59" y1="13.51" x2="15.42" y2="17.49"/><line x1="15.41" y1="6.51" x2="8.59" y2="10.49"/></svg>
             </div>
-            <h3>Built-in Classifiers</h3>
+            <h3>MLP Classifier</h3>
             <p>MLP classifier and Label Propagation included — pure numpy/scipy, no PyTorch, no GPU. Evaluate embedding quality directly without external dependencies.</p>
         </div>
         <div class="feature-card scroll-reveal" data-delay="100">
@@ -896,7 +896,7 @@ <h2>The Real Cost Per Embedding Job</h2>
     </div>
     <div class="cost-takeaway scroll-reveal">
         <div class="cost-takeaway-stat">&lt;$0.02</div>
-        <div class="cost-takeaway-text">to embed 2M nodes on the CPU machine (4.2 seconds at $13.10/hr). GPU methods need a $40.45/hr machine — <em>and</em> Cleora gives you more addressable memory (1 TB vs 640 GB VRAM). As your graph outgrows GPU VRAM, Cleora keeps running. GPU methods fail.</div>
+        <div class="cost-takeaway-text">to embed 2M nodes on the CPU machine (5.3 seconds at $13.10/hr). GPU methods need a $40.45/hr machine — <em>and</em> Cleora gives you more addressable memory (1 TB vs 640 GB VRAM). As your graph outgrows GPU VRAM, Cleora keeps running. GPU methods fail.</div>
     </div>
 </section>