Add retry support for infrastructure errors in vf-eval

[rasdani]

Description

Adds --max-retries CLI flag to retry rollouts when vf.InfraError occurs.
I've opted to only retry on vf.InfraError right now, as we don't assure yet that any vf.Error is retry-able.

Changes:

• async_utils.py: Add maybe_retry() utility using tenacity with exponential backoff + jitter
• types.py: Add max_retries to EvalConfig
• eval.py: Add --max-retries CLI argument (default: 0)
• eval_utils.py / environment.py: Thread parameter through to generate()

Behavior:

• Only retries vf.InfraError (transient infra failures)
• Does NOT retry ToolError, OverlongPromptError, etc.
• Logs retries at WARNING level

Usage:
vf-eval my-env --max-retries 3

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Test improvement

Testing

• All existing tests pass when running uv run pytest locally.
• New tests have been added to cover the changes

Checklist

• My code follows the style guidelines of this project as outlined in AGENTS.md
• 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

---

Note

Adds configurable retries for transient infrastructure failures.

• Introduces maybe_retry() in utils/async_utils.py (tenacity-based exponential backoff + jitter) and applies it to Environment.generate() by wrapping run_rollout/run_group
• Threads max_retries through Environment.generate/evaluate/evaluate_sync, EvalConfig (verifiers/types.py), and run_evaluation
• Extends CLI with --max-retries in verifiers/scripts/eval.py; default 0
• Updates docs (docs/evaluation.md, docs/reference.md) to document the flag and config fields
• Adds tests for retry behavior and CLI defaults (tests/test_environment.py, tests/test_eval_cli.py)

^{Written by Cursor Bugbot for commit adc524a. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}

[mikasenghaas]

nice! this looks clean, 2 minor comments