feat: komplette Codebase vom Board ins Repo gespiegelt

app/ = /opt/callrec (Node-Backend, Vite-Frontend, Python-Inferenz,
Audio-Skripte, rkllm_server.cpp), system/ = systemd-Units,
WirePlumber/PipeWire-Drop-ins, sudoers. Modelle, Binaries, Aufnahmen
und node_modules bleiben draussen (.gitignore).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

app/inference/smoke_npu.py

@@ -0,0 +1,23 @@
+import sys, time
+import numpy as np
+sys.path.insert(0, "/opt/callrec")
+from inference.rknpu_whisper_encoder import RknnWhisperEncoder
+
+M = "/opt/callrec/models/turbo-encoder-fp16.rknn"
+print("== init ==", flush=True)
+t0 = time.time()
+e = RknnWhisperEncoder(M, lib_path="/usr/lib/librknnrt.so")
+print(f"init OK in {time.time()-t0:.1f}s", flush=True)
+
+rng = np.random.default_rng(9)
+mel = (rng.standard_normal((1,128,3000))*0.5).astype(np.float32)
+print("== run ==", flush=True)
+t0 = time.time()
+out = e.run(mel)
+dt = time.time()-t0
+print(f"run OK in {dt:.2f}s  out.shape={out.shape} dtype={out.dtype}", flush=True)
+print(f"out mean={out.mean():.4f} std={out.std():.4f} min={out.min():.3f} max={out.max():.3f}", flush=True)
+# second run (warm) timing
+t0=time.time(); e.run(mel); print(f"warm run {time.time()-t0:.2f}s", flush=True)
+e.close()
+print("== done ==", flush=True)