Normalize multiple choice responses like "0: Yes" to code-only format

[Fizza-Mukhtar]

This PR addresses issue #2356.

Problem:

• When use_code=True for multiple choice questions, the LLM sometimes returns responses like "0: Yes" instead of just "0".
• This causes QuestionAnswerValidationError because the strict literal validation expects only the numeric code.

Solution:

• Added normalization in MultipleChoiceResponseValidator.fix() to split responses containing ":" and keep only the numeric code.
• Applied only for string responses when use_code=True.
• Existing validation and matching strategies (exact, case-insensitive, prefix matching) remain unchanged.

This ensures that the validator accepts responses like "0: Yes" while keeping the validation strict for other cases.