High-quality instruction-following (IF) evaluation datasets in VERL format: verifiable constraints and constraint-based benchmarks for RL training