[FEAT] Support both multiple-choice and single-answer card generation

[MikeNovikoff]

Feature description

The AI generation pipeline must support two pedagogical modes: standard multiple-choice quizzes (with potentially multiple correct answers) and open-ended single-answer questions.

What is needed to be done

    llm-service/generator.py: Two Pydantic response schemas (MultipleChoiceList, SingleAnswerList)
    llm-service/generator.py: Dynamic prompt selection based on card_type
    quiz_core/internal/api/handlers/llm/generate.go: Validate card_type enum

Acceptance Criteria

• Prompt for multiple_choice enforces exactly 4 options, no prefix labels (A), B), etc.), and 0-based correct_indices
• Prompt for single_answer enforces a concise correct_answer string
• Both modes include a short hint for each card
• quiz_core rejects invalid card_type values with 400 Bad Request