Don't show the live "recording"/listening indicator until the model is ready

startLive()'s browser-engine branch flipped liveStatus to "listening" (the
red live-dot, "Listening..." hint) the instant the mic was granted, even if
the Whisper pipeline was still loading — the user saw an active-looking
session while nothing was being transcribed yet. transcribeLiveSegment()
already correctly awaits the model-ready gate before posting a segment, so
this was purely a misleading-status bug.

Adds a "loading-model" LiveStatus between "starting" and "listening";
startLive() now waits on the same browserModelReady() gate the transcribe
path uses before claiming "listening", with race-safe cleanup if the load
fails while waiting. Server-engine live is unaffected.

Adds Decisions.md D-034; updates tests/live.md's manual verification steps.

src/App.tsx

@@ -889,8 +889,34 @@ export function App() {
 
     try {
       const mic = await startMicCapture((chunk) => transcriber.push(chunk));
-      liveRef.current = { mic, transcriber };
-      setLiveStatus("listening");
+      const session = { mic, transcriber };
+      liveRef.current = session;
+      if (loadProgress.ready) {
+        // Pipeline already loaded for the current model/device — nothing to
+        // wait on, go straight to "listening" as before.
+        setLiveStatus("listening");
+      } else {
+        // Model not ready yet (first use, or a model/device switch): don't
+        // claim "Listening" until the pipeline actually is one, otherwise the
+        // live-dot/hint imply an active session while nothing is being
+        // transcribed (D-034). `transcribeLiveSegment` already awaits the
+        // same gate before posting, so no segments are lost either way.
+        setLiveStatus("loading-model");
+        try {
+          await browserModelReady();
+          // The session could have been torn down (Stop, or another start)
+          // while we were waiting — only advance if it's still this one.
+          if (liveRef.current === session) setLiveStatus("listening");
+        } catch {
+          // Load failed; the worker's own "error" handler already calls
+          // setError() for stage:"load" failures, so don't duplicate it here.
+          if (liveRef.current === session) {
+            liveRef.current = null;
+            await mic.stop().catch(() => {});
+            setLiveStatus("idle");
+          }
+        }
+      }
     } catch (micError) {
       setLiveStatus("idle");
       const message =