### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

The Noma Security v2 guardrail (noma_v2, also reachable as guardrail: noma with use_v2: true) does not honor the shared streaming knobs streaming_end_of_stream_only and streaming_sampling_rate. Setting either in config.yaml has no effect: streaming post-call scans always run per chunk at the hard-coded default sampling rate of 5. There is no error and no warning, so the misconfiguration is invisible.

NomaV2Guardrail implements apply_guardrail and does not define its own async_post_call_streaming_iterator_hook. litellm/proxy/utils.py therefore classifies it as an apply_guardrail callback and routes its streaming post-call through UnifiedLLMGuardrails.async_post_call_streaming_iterator_hook:

litellm/proxy/utils.py:

if "async_post_call_streaming_iterator_hook" in cls_attrs: has_iterator_override = True iterator_overrides.append((resolved, "override")) elif "apply_guardrail" in cls_attrs: iterator_overrides.append((resolved, "apply_guardrail"))

That dispatcher resolves the knobs off the guardrail instance with getatt…