feat(eval): add evaluation runtime tracking and detailed state metadata

[srthkdev]

Enhanced Metadata Export in vf-eval

Description

This PR enhances the metadata export functionality in the vf-eval CLI tool to provide more comprehensive information when running evaluations with the -s flag. The additional metadata will be useful for analysis and display on the hub. solves #307

Key Changes

1. Runtime Tracking: Added evaluation runtime tracking with eval_runtime_seconds field in metadata
2. Parser Results: Added parsed_answer field containing results from Parser.parse_answer()
3. Rubric State Metadata: Added state_metadata field containing:
   • Judge responses from rubric state
   • Model responses from API calls
   • Tool call information
   • Other custom serializable state fields
4. Additional Metrics: Added total_rollouts field to track total number of rollouts performed

Type of Change

• New feature (non-breaking change which adds functionality)

Testing

• All existing tests pass
• Code compiles successfully without syntax errors
• Enhanced metadata fields are properly structured and exported

Test Coverage

• Current coverage: Not changed
• Coverage after changes: Not changed

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• Any dependent changes have been merged and published

Additional Notes

Potential Additional Enhancements

More Detailed Performance Metrics:

• Add breakdown of time spent in different phases (API calls, parsing, scoring)
• Add throughput metrics (rollouts per second)

Environment-Specific Metadata:

• Extract and export environment-specific configuration
• Include dataset information (size, source, etc.)

Model Response Analysis:

• Add token usage statistics from model responses
• Include finish reasons from model responses

Enhanced Error Tracking:

• Track and export information about failed rollouts
• Include error types and counts

System Information:

• Export system information (Python version, library versions)
• Include hardware information when available