feat: OpenVINO CPU MetricGAN mask; 100 ms hop

Per-worker OpenVINO CPU runs the MetricGAN BLSTM mask; SpeechBrain
keeps STFT/ISTFT. ENHANCE_INFER=auto with torch fallback. Hop is
0.1 s after isolated last_ms ~6 ms (model+ov+beamform).

run.py

@@ -46,6 +46,7 @@ def worker_args_for(cfg, cam) -> list[str]:
         "--enhance-model", cfg.enhance_model,
         "--enhance-chunk-s", str(cfg.enhance_chunk_s),
         "--enhance-hop-s", str(cfg.enhance_hop_s),
+        "--enhance-infer", cfg.enhance_infer,
         "--beamform-mode", cfg.beamform_mode,
         "--torch-num-threads", str(cfg.torch_num_threads),
         "--tdoa-every", str(cfg.beamform_tdoa_every),