fix(ai): revert schema to optional and explicitly instruct LLM to pass null for Groq tool compatibility

Author: yash-pouranik

apps/python-service/routers/ai.py

@@ -59,7 +59,7 @@ class CollectionCreatorRequest(BaseModel):
 class CollectionCreatorResponse(BaseModel):
     type: Literal["clarify", "schema", "complete"]
     message: str
-    schema_: list[CollectionSchema] = Field(..., alias="schema", description="Must be an empty array [] if type is 'clarify'")
+    schema_: list[CollectionSchema] | None = Field(default=None, alias="schema", description="MUST be null if type is 'clarify' or 'complete'")
 
     model_config = ConfigDict(populate_by_name=True)
 
@@ -169,7 +169,7 @@ async def collection_creator(request: CollectionCreatorRequest, req: Request):
         system_prompt = """You are a MongoDB schema designer for urBackend, a Backend-as-a-Service platform.
 
 Rules:
-1. If the description is vague, ask 2-3 specific clarifying questions and set type: "clarify". Set schema: [] to satisfy the tool schema. NEVER ask more than 3 at once.
+1. If the description is vague, ask 2-3 specific clarifying questions and set type: "clarify". You MUST set "schema": null. NEVER ask more than 3 at once.
 2. Once you have enough context, propose a schema and set type: "schema".
 3. Only use these field types: String, Number, Boolean, Date, Object, Array, Ref.
 4. ALWAYS include "createdAt" (type: Date, required: true) in EVERY collection.