Adds python scripts to support curriculum learning experiments

[sruan2]

Summary

This PR adds three python scripts to support curriculum learning experiments in AppWorld.

Changes

Added three new scripts in experiments/curriculum/:

analyze_dataset.py - Task difficulty distribution analyzer
Saves analysis results to JSON files in the same dataset folder

data_selector.py - Curriculum dataset builder
Selects subsets of tasks from existing datasets based on size, difficulty, and a range of requirements specified

check_same_data.py - Dataset comparison utility
Compares two dataset files to verify they contain the same tasks

[sruan2]

Two additional changes:

1. Fixed SN endpoint timeouts by adding error handling and increasing the timeout
   experiments/code/ace/lite_llm_generator.py

2. Added an embedding-based clustering method using the OpenAI embeddings endpoint
   experiments/curriculum/similarity_metrics.py