Bug: Gemini realtime path returns empty success (score=1.0, output=None) on a safety-blocked or refused response

[mittalpk]

Describe the overall issue and situation

In the Gemini provider's realtime (non-batch) inference path, a safety-blocked prompt, a candidate that stopped for a non-STOP reason (safety, recitation, etc.) with no content, or a response containing only non-text parts is returned as a successful, empty extraction instead of raising.

langextract/providers/gemini.py (_process_single_prompt):

response = self._client.models.generate_content(
    model=self.model_id, contents=prompt, config=call_config
)
return core_types.ScoredOutput(score=1.0, output=response.text)

google.genai's GenerateContentResponse.text (_get_text() in the SDK) returns None, not an exception, when:

• not self.candidates — the prompt itself was blocked (prompt_feedback.block_reason set),
• candidates[0].content or .content.parts is empty — the candidate stopped for a non-STOP finish_reason (SAFETY, RECITATION, PROHIBITED_CONTENT, etc.) with no output, or
• the response contains only non-text parts (e.g. a function call).

ScoredOutput.output is typed str | None, so None is accepted and returned with score=1.0. Downstream, this is reported as a successful empty extraction — the caller gets no signal that anything was blocked, and the block/refusal reason (prompt_feedback.block_reason, candidates[0].finish_reason) is silently discarded.

This is the same class of bug reported for the OpenAI realtime path in #491 (fixed for OpenAI in #496). #491 explicitly noted: "The Gemini realtime path has the same class of gap on a safety-blocked response.text (version-dependent); worth folding the same guard in there too." PR #496 only touched openai.py, so this side was never addressed.

Expected behavior

When response.text is None, _process_single_prompt raises InferenceRuntimeError with the available diagnostic (block reason / finish reason), mirroring the fix already applied to the OpenAI provider.

Actual behavior

_process_single_prompt returns ScoredOutput(score=1.0, output=None). The caller receives a "successful" extraction with no content; the actual block/refusal reason is lost.

Steps to reproduce the issue

1. Configure GeminiLanguageModel and issue a prompt that trips Gemini's safety filters, or otherwise produces a candidate with a non-STOP finish reason and no text content.
2. The SDK's response.text evaluates to None (confirmed directly against google-genai 2.11.0's GenerateContentResponse._get_text).
3. Observe _process_single_prompt returns ScoredOutput(score=1.0, output=None) instead of raising — the extraction is reported as a successful empty result.

Any additional content

• Verified without live API: mock models.generate_content to return a response with text=None, prompt_feedback.block_reason or candidates[0].finish_reason set, and assert _process_single_prompt raises InferenceRuntimeError surfacing the reason (mirrors the existing OpenAI refusal test added in Handle OpenAI refusals and missing message content #496).
• Environment: current main at commit b5fe0ba.