feat: Handle empty Ollama responses gracefully

Introduce `OllamaError::EmptyResponse` to represent cases where the
Ollama
model returns an empty result due to silence rules or lack of keywords.

This change prevents treating an empty response as a parse error and
logs
it at a lower severity level. A new integration test verifies this
behavior.

server/tests/transcribe_test.rs

@@ -500,6 +500,31 @@ async fn ollama_oneliner_sends_keep_alive_and_model() {
     // Mock::expect(1) + server drop verifies the matcher hit exactly once.
 }
 
+#[tokio::test]
+async fn ollama_oneliner_empty_response_when_model_obeys_silence_rule() {
+    // Model followed the silence rule for a keyword-less transcript and
+    // returned empty content. That's a valid outcome, not a parse error.
+    let server = MockServer::start().await;
+    Mock::given(method("POST"))
+        .and(path("/api/chat"))
+        .respond_with(ResponseTemplate::new(200).set_body_json(ok_chat_body("   \n  ")))
+        .mount(&server)
+        .await;
+
+    let client = reqwest::Client::new();
+    let err = generate_oneliner(
+        &client,
+        &server.uri(),
+        MODEL,
+        0,
+        "äh, Moment",
+        Duration::from_secs(5),
+    )
+    .await
+    .expect_err("expected EmptyResponse");
+    assert!(matches!(err, OllamaError::EmptyResponse), "got {err:?}");
+}
+
 #[tokio::test]
 async fn ollama_oneliner_parse_error_on_missing_content() {
     let server = MockServer::start().await;