What

• Detect and handle LLM provider refusal responses (e.g., Anthropic's stop_reason: "refusal") in SDK1's LLM.complete(), acomplete(), and stream_complete() methods.
• Raise a descriptive LLMError instead of crashing with 'NoneType' object has no attribute 'find'.

Why

• When Anthropic (or other providers) refuse to generate a response due to safety restrictions, LiteLLM returns content=None with finish_reason="refusal". The SDK did not check for None content before passing it to _post_process_response(), which called .find() on None.
• This produced a confusing, non-descriptive error: "Error from LLM provider 'anthropic': 'NoneType' object has no attribute 'find'" — making it very hard to diagnose the actual issue (a safety refusal).

How

• llm.py — complete() and acomplete(): After extracting response_text from the LiteLLM response, check if it's None. If so, inspect finish_reason and raise LLMError with a clear message.
• llm.py — stream_complete(): Extracted chunk processing into _process_stream_chunk() helper. Checks finish_reason on each chunk for refusal.
• llm.py — _raise_for_empty_response(): New helper method that raises LLMError with status 400 for refusals, 500 for other empty responses.
• common.py — LLMResponseCompat: Hardened __str__() and __repr__() to handle None text without crashing.

Before: "Error from LLM provider 'anthropic': 'NoneType' object has no attribute 'find'"
After: "The LLM refused to generate a response due to safety restrictions. Please review your prompt and try again."

Can this PR break any existing features. If yes, please list possible items. If no, please explain why. (PS: Admins do not merge the PR without this section filled)

• No. This only adds an early check for None content before existing processing logic runs. Normal responses (where content is a string) are completely unaffected — the new check is skipped. The only behavioral change is that refusal/empty responses now raise a descriptive LLMError instead of an opaque AttributeError.

Database Migrations

• None

Env Config

• None

Relevant Docs

• Anthropic: Handling Stop Reasons — documents stop_reason: "refusal"
• LiteLLM: Anthropic Provider — LiteLLM passes "refusal" through as finish_reason

Related Issues or PRs

• N/A

Dependencies Versions

• No new dependencies

Notes on Testing

• Tested with mocked LiteLLM responses simulating refusal (content=None, finish_reason="refusal"), empty responses, and normal responses. All 13 new tests pass alongside the existing 70 SDK1 tests (83 total).

Checklist

I have read and understood the Contribution Guidelines.