Fix (1,T) shape crash in _cf_join; bump HunyuanFoley secs_per_step to 0.5

- _load_seg_wavs: squeeze (1, T) → (T,) on load. HunyuanFoley's DAC
decoder emits (batch, 1, time); audio_batch[0] = (1, T). When these
single-channel arrays are saved as npy and later loaded alongside
true stereo (2, T) from MMAudio, _cf_join tries to concatenate
(2, N) with (1, M) along axis=1 → ValueError. The squeeze at load
time normalises all single-channel wavs to plain mono (T,) so
_resample_to_slot_sr's channel-matching logic works correctly for
cross-model regens and stitching.

- HunyuanFoley secs_per_step: 0.35 → 0.50. Measured wall time from
logs was 0.607 s/step (includes VAE decode + feature extraction per
segment, not just the denoising loop). More accurate estimate avoids
under-reserving ZeroGPU duration for multi-segment videos.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

app.py

@@ -328,8 +328,19 @@ def _save_seg_wavs(wavs: list[np.ndarray], tmp_dir: str, prefix: str) -> list[st
 
 
 def _load_seg_wavs(paths: list[str]) -> list[np.ndarray]:
-    """Load segment wav arrays from .npy file paths."""
-    return [np.load(p) for p in paths]
+    """Load segment wav arrays from .npy file paths.
+
+    Normalises (1, T) arrays → (T,) mono so that single-channel output from
+    models like HunyuanFoley (DAC decoder emits shape (1, T)) never causes a
+    shape mismatch in _cf_join when mixed with true stereo (2, T) arrays.
+    """
+    wavs = []
+    for p in paths:
+        w = np.load(p)
+        if w.ndim == 2 and w.shape[0] == 1:
+            w = w.squeeze(0)   # (1, T) → (T,)  mono
+        wavs.append(w)
+    return wavs
 
 
 # ------------------------------------------------------------------ #
@@ -552,7 +563,7 @@ MODEL_CONFIGS = {
     "hunyuan": {
         "window_s":       15.0,   # HunyuanFoley max video duration
         "sr":             48000,
-        "secs_per_step":  0.35,   # measured 0.328 s/step on H200
+        "secs_per_step":  0.50,   # wall-time ~0.607s/step (incl. VAE decode + feature extraction)
         "load_overhead":  90,     # cold disk: ~73s for 10 GB weights + ~8s aux models
         "tab_prefix":     "hf",
         "label":          "HunyuanFoley",