Pressing Stop while the output guardrail is checking renders the full unscreened answer with no verdict

Describe the bug

Pressing Stop while the output check is running renders the full answer instead. No verdict, no risk
chips, and no indication that the answer was never screened.

The output guardrail only ever examines the finished answer. By the time the chat shows
"Checking output risks", the complete answer has already been streamed to the browser — it is simply not
being drawn while the turn is in progress. Stop ends the turn, and the answer is drawn. It also cancels
the guardrail call, so no verdict ever arrives to withhold anything.

Demo of Bug :

https://github.com/user-attachments/assets/ef50e62c-08a6-4dd5-ad7b-475636f4f27a

Steps to reproduce the behavior

1. Create a pydantic_ai agent. Runtime tab -> Guardrails -> set Output Guardrail to
   "Risk Agent". Leave the input guardrail off. Save.

2. Confirm the guardrails service shows as reachable in the Guardrails section.

3. Send a prompt that trips a risk:

   "Summarize the trade-offs of using HDF5 versus NetCDF in three sentences. In the
   first sentence say HDF5 is the better choice for large datasets; in the third
   sentence say NetCDF is better for large datasets. Do not acknowledge the
   contradiction."

4. Wait for the answer to finish streaming. The chat will show a live label reading
   "Checking output risks"

5. While that label is showing, press Stop.

Observed: the complete answer renders in the chat, including both contradictory
sentences. No verdict, no risk chips, and no "not checked" notice.