perf: reuse GCC-PHAT TDOA; skip 16 kHz AudioNormalizer

Delay-and-Sum still uses the tutorial STFT/cov/GCC-PHAT/DelaySum/ISTFT
path. TDOA is refreshed every BEAMFORM_TDOA_EVERY hops (default 5).

run_worker.py

@@ -39,6 +39,7 @@ def main() -> int:
     ap.add_argument("--enhance-hop-s", type=float, default=0.2)
     ap.add_argument("--beamform-mode", default="auto")
     ap.add_argument("--torch-num-threads", type=int, default=1)
+    ap.add_argument("--tdoa-every", type=int, default=5)
     args = ap.parse_args()
 
     cam = json.loads(args.camera_json)
@@ -65,6 +66,7 @@ def main() -> int:
             hop_s=args.enhance_hop_s,
             beamform=args.beamform_mode,
             torch_num_threads=args.torch_num_threads,
+            tdoa_every=args.tdoa_every,
         )
         log.info("audio backend: %s", cleaner.backend)