Stable: decoder 8T optimization + restore pre-computed embeds

- BigVGAN: 8 threads (2757→1872ms), pre_conv/pre_transformer: 4 threads
- Restored pre-computed embeds format (codec_sum+text from Python)
- Text-only trailing embeds don't work: model needs codec_sum for EOS

For long phrases, pre-computed embeds must be generated from Python.
RTF 1.26 on short phrase.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

executorch-custom/jni_layer_tts.cpp

@@ -839,7 +839,8 @@ ExecuTorchJni::runTtsPipelineImpl(
         for(int i=0;i<NUM_CB;i++) allCodes.push_back(codes[i]);
         cb0Hist.push_back(curCb0);
 
-        // Next embed: pre-computed trailing OR codec_sum + eos/pad
+        // Next embed: use pre-computed trailing embeds (codec_sum+text from Python)
+        // then codec_sum + eos/pad after trailing exhausted
         float nextEmb[DIM]={};
         if(trIdx<nTrailing){
             memcpy(nextEmb,trailing.data()+trIdx*DIM,DIM*4);