Add a node to zero out the cond to advanced/conditioning

The stability streamlit example passes a zero cond as the negative input
so using this for the negative input makes outputs match the streamlit.

nodes.py

@@ -148,6 +148,25 @@ class ConditioningSetMask:
             c.append(n)
         return (c, )
 
+class ConditioningZeroOut:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {"conditioning": ("CONDITIONING", )}}
+    RETURN_TYPES = ("CONDITIONING",)
+    FUNCTION = "zero_out"
+
+    CATEGORY = "advanced/conditioning"
+
+    def zero_out(self, conditioning):
+        c = []
+        for t in conditioning:
+            d = t[1].copy()
+            if "pooled_output" in d:
+                d["pooled_output"] = torch.zeros_like(d["pooled_output"])
+            n = [torch.zeros_like(t[0]), d]
+            c.append(n)
+        return (c, )
+
 class VAEDecode:
     @classmethod
     def INPUT_TYPES(s):
@@ -1350,6 +1369,8 @@ NODE_CLASS_MAPPINGS = {
 
     "LoadLatent": LoadLatent,
     "SaveLatent": SaveLatent,
+
+    "ConditioningZeroOut": ConditioningZeroOut,
 }
 
 NODE_DISPLAY_NAME_MAPPINGS = {