feat(eval): add support for grouped reward summaries and reports

[srthkdev]

• Add command-line options to group rewards by task or custom keys
• Implement grouped summary computation for evaluation results
• Log grouped reward and metric statistics during evaluation
• Enhance HTML report to show grouped reward summaries by specified keys
• Add grouped metric calculations in GRPOTrainer for detailed analysis
• Ensure backward compatibility with default grouping by task
• Handle exceptions in HTML report generation gracefully

Soon, I will add screenshots solves #298 testing, open for review.

[willccbb]

Ah we deprecated the report feature, sorry!