Add data analysis scripts and figure generation pipeline

[patelfagun1998]

Summary

• Add paper figure scripts: per-attempt accuracy, Kendall tau decomposition, behavioral comparison, error type flow, response matrix visualization, learning curves, dataset summary table
• Add scripts/build_figures.sh to orchestrate all figure generation and copy to overleaf
• Add supporting analysis: LLM testcase analysis, student behavior clustering refactor, problem-by-problem and pace analysis updates
• Add scripts/plot_student_model_comparison.py and scripts/prototype_student_comparison.py
• Add rssm_full_adapter.py and llm_diagnostics.py (deferred from earlier PRs)

Paper figures produced

Script	Figure
plot_filtered_accuracy.py	accuracy_vs_attempt (Section 6)
llm_prediction_deep_analysis.py	decomposition_test (Section 6)
llm_behavioral_comparison.py	behavioral_combined (Section 6)
llm_bug_comparison.py	error_type_flow (Section 6)
visualize_response_matrix.py	attempt_progression_compact, avg_improvement_by_attempt (Section 6)
learning_curves_by_year.py	learning_curves_combined (Section 3)
dataset_summary_table.py	tab:dataset_summary (Section 3)
student_behavior_clustering.py	student_behavior_all_courses (Appendix A10)
problem_by_problem_analysis.py	aggregate_problem_patterns (Appendix A10)
pace_analysis.py	aggregate_submission_patterns (Appendix A10)

Test plan

• bash scripts/build_figures.sh — generates all figures and copies to overleaf/figures/
• Verify output figures exist in overleaf/figures/