fix: improve JSON parsing resilience in synthesis module

[himanshu231204]

Summary

• Enhance _parse_response() in QuestionGenerator to handle malformed JSON from LLM
• Add JSON array extraction from responses with extra text before/after
• Fix trailing commas before ] or } in LLM JSON output
• Replace single quotes with double quotes when no double quotes present
• Apply same fixes to AdversarialTestCaseGenerator
• Add 10 new tests covering trailing commas, extra text, single quotes, markdown code blocks, and multiple items
• All 31 tests pass

Changes

• openagent_eval/synthesis/question_gen.py: Enhanced _parse_response() with regex extraction and JSON fixes
• openagent_eval/synthesis/adversarial.py: Applied same JSON resilience fixes
• ests/unit/test_synthesis/test_question_gen.py: Added 5 new tests for edge cases
• ests/unit/test_synthesis/test_adversarial.py: Added 5 new tests for edge cases
• examples/corpus_and_related_modules.ipynb: Added demo notebook using real Groq LLM