feat(app): speak LLM reply via Windows SAPI server-side

GUI front-end does not subscribe to IpcEvent::LlmReply, so the LLM
auto-fallback was effectively silent — user got the OK chime but no
spoken answer. Backlog item resolved.

llm_fallback::handle now also invokes speak_via_sapi() after firing
IpcEvent::LlmReply (both for successful replies and the fallback
error message). The text goes through System.Speech.Synthesis.
SpeechSynthesizer, with a preference pass that picks the first
installed ru-RU voice if present (Microsoft Irina / Pavel etc.) so
Russian replies sound right out of the box, and falls back to the
default English voice otherwise.

PowerShell is invoked fire-and-forget (no Wait, stdout/stderr null)
so the voice loop never blocks. A long answer keeps speaking while
jarvis returns to wake-word listening; the mic may pick up some of
the speech which is the same compromise voices::play_* already makes.
A cleaner version would pause pv_recorder for the duration of the
utterance — that goes in the backlog as part of the eventual
unified IpcEvent::Speak refactor.

Toggle: set JARVIS_LLM_TTS=false to disable (e.g. on a server where
the GUI is the speaker). Non-Windows builds get a no-op stub.

crates/jarvis-app/src/llm_fallback.rs

@@ -98,16 +98,50 @@ pub fn handle(prompt: &str) {
             let reply = reply.trim().to_string();
             info!("LLM reply: {}", reply);
             state.history.lock().push_assistant(reply.clone());
-            ipc::send(IpcEvent::LlmReply { text: reply });
+            ipc::send(IpcEvent::LlmReply { text: reply.clone() });
             voices::play_ok();
+            speak_via_sapi(&reply);
         }
         Err(e) => {
             error!("LLM request failed: {}", e);
             state.history.lock().pop_last_user();
-            ipc::send(IpcEvent::LlmReply {
-                text: config::LLM_FALLBACK_ERROR_RU.to_string(),
-            });
+            let err_text = config::LLM_FALLBACK_ERROR_RU.to_string();
+            ipc::send(IpcEvent::LlmReply { text: err_text.clone() });
             voices::play_error();
+            speak_via_sapi(&err_text);
         }
     }
 }
+
+#[cfg(target_os = "windows")]
+fn speak_via_sapi(text: &str) {
+    if std::env::var("JARVIS_LLM_TTS").as_deref() == Ok("false") {
+        return;
+    }
+    let escaped = text.replace('\'', "''");
+    let ps = format!(
+        "Add-Type -AssemblyName System.Speech; \
+         $s = New-Object System.Speech.Synthesis.SpeechSynthesizer; \
+         foreach ($v in $s.GetInstalledVoices()) {{ \
+           if ($v.VoiceInfo.Culture.TwoLetterISOLanguageName -eq 'ru') {{ \
+             try {{ $s.SelectVoice($v.VoiceInfo.Name); break }} catch {{ }} \
+           }} \
+         }} \
+         $s.Speak('{}')",
+        escaped
+    );
+    match std::process::Command::new("powershell")
+        .args(["-NoProfile", "-Command", &ps])
+        .stdout(std::process::Stdio::null())
+        .stderr(std::process::Stdio::null())
+        .spawn()
+    {
+        Ok(_) => {}
+        Err(e) => warn!("SAPI spawn failed: {}", e),
+    }
+}
+
+#[cfg(not(target_os = "windows"))]
+fn speak_via_sapi(_text: &str) {
+    // No-op on non-Windows; LLM reply still arrives via IPC for the GUI to render.
+}