fix(web): reduce FOT by stripping redundant voice data from transport

Three transport-only optimizations that cut per-session Vercel FOT by ~50-60%:

P0 — Server strips voice.referenceAudioBase64 from already-known characters
in /api/scene and /api/insert-beat responses (defense-in-depth).

P1 — Client strips all voice data from session before sending to
/api/scene, /api/vision, and /api/insert-beat. Voices are retained locally
and re-merged from responses via mergeCharactersPreserveVoice(). The engine
only needs character names + visualDescriptions for scene generation.

P3 — /api/beat-audio returns binary audio (Response with Content-Type)
instead of JSON-wrapped base64, saving ~33% encoding overhead. Client
converts to blob URLs; PlayCanvas accepts a single audioSrc prop.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

app/api/insert-beat/route.ts

@@ -26,7 +26,10 @@ export async function POST(req: Request) {
     // See StartRequest.clientTts — BYO clients synth in-browser, so drop server TTS.
     const config = body.clientTts === true ? { ...base, tts: undefined } : base;
     const result = await requestInsertBeat(config, body);
-    return NextResponse.json(result);
+    return NextResponse.json({
+      ...result,
+      characters: result.characters.map((c) => ({ ...c, voice: undefined })),
+    });
   } catch (err) {
     const message = err instanceof Error ? err.message : "Unknown error";
     return NextResponse.json({ error: message }, { status: 500 });