docs: add SaveEmbedder/LoadEmbedder to README and landing page

The v1.2.0 release introduced embedder state persistence
(rp_persistence.go) but the feature was undocumented in both
the README and the website. This closes the gap:

README.md:
  - New 'Embedder Persistence' subsection under Persistence
    with full Go example (fit → save → load → embed)
  - Added SaveEmbedder / LoadEmbedder to the RandomProjections
    API bullet list

docs/index.html:
  - New 'Embedder state persistence' card in the Persistence
    section with syntax-highlighted code block

Author: molty3000

README.md

@@ -61,6 +61,8 @@ The `Embedder` interface lets you swap backends: bring your own OpenAI, Ollama,
 
 ## Persistence
 
+### Store Persistence
+
 ```go
 // Save to disk (gob — compact binary)
 store.Save("/data/vectors.db")
@@ -75,6 +77,34 @@ restored.Load("/data/vectors.db")
 
 Gob-encoded stores are compact (~4 bytes per float32 + overhead). For a 10K × 1536d store, expect ~60 MB on disk and ~200ms save/load times.
 
+### Embedder Persistence
+
+```go
+// Fit an embedder on your corpus
+rp := vector.NewRandomProjections(256)
+rp.Fit([]string{
+    "machine learning is fascinating",
+    "deep neural networks transform AI",
+    "the weather today is sunny",
+})
+
+// Save the embedder state — vocab, projection matrix, dimensions
+if err := rp.SaveEmbedder("/data/embedder.gob"); err != nil {
+    log.Fatal(err)
+}
+
+// Later — load without refitting
+restored, err := vector.LoadEmbedder("/data/embedder.gob")
+if err != nil {
+    log.Fatal(err)
+}
+
+// Embeddings are deterministic — same text, same vector
+v, _ := restored.Embed("machine learning")
+```
+
+`SaveEmbedder` preserves the vocabulary, projection matrix, output dimension, and scaling factor. `LoadEmbedder` reconstructs a ready-to-use `*RandomProjections` that produces identical vectors to the original.
+
 ## API
 
 ### Vector Type
@@ -139,6 +169,8 @@ Johnson-Lindenstrauss sparse random projection (Achlioptas 2003). Projects token
 - `Embed(text string) (Vector, error)` — embed text (L2-normalized output)
 - `VocabSize() int` — number of unique tokens in vocabulary
 - `Dims() int` — output dimensionality
+- `SaveEmbedder(path string) error` — persist embedder state to gob file
+- `LoadEmbedder(path string) (*RandomProjections, error)` — restore embedder from gob file
 
 ## Performance
 

docs/index.html

@@ -372,6 +372,17 @@ <h3>JSON — human-readable</h3>
       <div class="code-mini"><pre>store.<span class="fn">SaveJSON</span>(<span class="str">"/data/vectors.json"</span>)
 restored.<span class="fn">LoadJSON</span>(<span class="str">"/data/vectors.json"</span>)</pre></div>
     </div>
+
+    <div class="usecase">
+      <div>
+        <h3>Embedder state persistence</h3>
+        <p>Save and restore the <code>RandomProjections</code> embedder — vocabulary, projection matrix, dimensions. No refitting needed across restarts.</p>
+      </div>
+      <div class="code-mini"><pre>rp.<span class="fn">SaveEmbedder</span>(<span class="str">"/data/embedder.gob"</span>)
+
+restored, _ := <span class="pkg">vector</span>.<span class="fn">LoadEmbedder</span>(<span class="str">"/data/embedder.gob"</span>)
+<span class="cmt">// Same text → identical vector</span></pre></div>
+    </div>
   </div>
 </section>