Remove debug prints from vision tower (fl2va passes e2e smoke)

src/h3_blackwell_runtime/qwen3vl_vision.py

@@ -332,7 +332,7 @@ class _VisionPatchMerger(nn.Module):
         # x: [t*h*w, hidden] (unmerged patches) for the main merger;
         #    [t*(h//2)*(w//2), merge_dim] (pre-merged 2x2) for the deepstack merger.
         if x.shape[-1] == self.merge_dim:
-            # Already 2x2-merged (deepstack style): norm over merge_dim.
+            # Deepstack merger: input is already 2x2-merged; norm over merge_dim.
             x = F.layer_norm(x, (self.merge_dim,), weight=self.norm_weight, bias=self.norm_bias, eps=1e-6)
         else:
             # Main merger: per-patch LayerNorm over hidden, then group 2x2 into merge_dim.