fix(envs): enhance error handling and add timeout guidance

[srthkdev]

• Add detailed timeout-specific guidance messages in environment model response errors
• Wrap rollout execution in try-except to log and propagate errors properly
• Improve AsyncBatchGenerator timeout checks with cleanup and informative error messages
• Wrap batch generation calls with error handling and ensure is_generating flag is cleared
• Extend evaluation timeout error messages with troubleshooting recommendations
• Provide consistent actionable suggestions to reduce max_concurrent, increase timeouts, check model server health, or adjust system limits across multiple components

Description

This PR enhances error handling and timeout management in the Verifiers library to address issue #103. The changes focus on providing more informative error messages with actionable guidance when timeout errors occur during training and evaluation.

Key improvements include:

• Enhanced timeout error messages with specific troubleshooting suggestions
• Better error handling in environment model responses
• Improved async batch generation timeout handling with cleanup
• More informative evaluation timeout error messages

All changes align with the reviewer's feedback to focus only on timeout handling and logging improvements while removing problematic functionality like automatic error filtering and brittle optimal training configuration.

Type of Change

• Bug fix (non-breaking change which fixes an issue)

Testing

• All existing tests pass
• New tests have been added to cover the changes
• Tests have been run locally with python -m pytest tests/

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

The changes in this PR specifically address the timeout error handling issues reported in issue #103. When users encounter timeout errors during training or evaluation, they will now receive detailed guidance on how to resolve the issues, including:

• Reducing max_concurrent parameter
• Increasing async_generation_timeout in GRPOConfig
• Verifying vLLM server is running and responsive
• Considering reducing max_tokens or using a smaller model
• Increasing system limits with 'ulimit -n 4096'

This PR does not include any of the previously discussed but rejected functionality like automatic error filtering or optimal training configuration, as requested by the reviewer #261 .