v0.2.0

Added

• Introduced a configuration-driven benchmarking module for evaluating
  backtracking performance.
• Integration with lm-evaluation-harness for running standardized NLP
  benchmarks.
• Hyperparameter optimization using optuna to find the best Operator
  settings.
• New CLI script for launching benchmark runs from a YAML configuration file.

Fixed

• Corrected the stop sequence handling to robustly and automatically stop on
  both the model's EOS token and user-defined sequences, preventing premature
  generation termination.