feat(question_gen): add v2 generator with per-family prompt templates

Implement generator_v2.py with:
- CandidateQuestion dataclass (canonical location)
- _load_prompt_template: loads per-family .md from store/prompts/
- _build_v2_prompt: constructs system+user messages with material context
- _parse_v2_response: JSON extraction, json_repair, field validation
- generate_one_v2: async VLM call orchestration with reject_reason support

Add 5 family-specific prompt templates:
- retrieval.md: factual recall from visible content
- reasoning.md: multi-hop inference across segments
- enumeration.md: counting/listing entities and actions
- visual.md: visual details requiring frame observation
- spatial.md: spatial relationships between objects/people

Tests: 11 unit tests covering prompt build, parse, and e2e generation.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

store/prompts/question_gen/visual.md

@@ -0,0 +1,22 @@
+You are a question generator for video understanding benchmarks.
+
+Your task: Generate a **visual detail** multiple-choice question that requires observing specific visual information from the video frames — details that cannot be answered from subtitles or text alone.
+
+## Guidelines
+
+- The question MUST target visual details: colors, shapes, positions, appearances, visual states, or visual actions.
+- The answer should be verifiable ONLY by looking at the actual frames — subtitle text alone must NOT suffice.
+- Focus on concrete visual observations: what objects look like, their appearance, visual relationships.
+- Distractors should be visually plausible alternatives that could be confused without careful observation.
+
+## Quality Requirements
+
+- Question must be grammatically correct and unambiguous.
+- All four options must be parallel in structure and length.
+- The correct answer must not be identifiable from linguistic cues alone.
+- Avoid questions about things that are typically described in subtitles (dialogue content, narration).
+- Each option must begin with "A. ", "B. ", "C. ", or "D. ".
+
+## Output
+
+Respond with ONLY a valid JSON object. No additional text.