#33086 [Bug]: Bedrock guardrail native hooks don't scan Anthropic tool_result content blocks
#### What happened?
`aws-bedrock-guardrails`' native proxy hooks (`async_pre_call_hook` / `async_moderation_hook` in `litellm/proxy/guardrails/guardrail_hooks/bedrock_guardrails.py`) silently skip PII/content scanning for Anthropic `tool_result` content blocks on live proxy traffic.
This is distinct from #29593 / #29594: those fix the same class of bug in the *Generic Guardrail API* translation layer (`litellm/llms/anthropic/chat/guardrail_translation/handler.py`), used by presidio/lakera/aim and by Bedrock's `/guardrails/apply_guardrail` testing endpoint. `BedrockGuardrail` sets `use_native_during_call_hook = True`, so its live `async_pre_call_hook`/`async_moderation_hook` path never goes through that translation layer at all — it has its own independent extraction logic with the identical bug, untouched by #29594.
#### Root cause
In `get_content_items_for_message` (`litellm/proxy/guardrails/guardrail_hooks/bedrock_guardrails.py`), list-content items are only recognized as text if they carry a top-level `"text"` key:
```python elif isinstance(content, list): for item in content: if isinstance(item, dict) and "text" in item: ... ```
An Anthropic `tool_…