feat: add workspace setup and teardown scripts

[christso]

Summary

Add support for custom setup and teardown scripts that run before/after each eval case when using workspace_template.

Background

PR #180 adds workspace_template which copies a directory to a temp location for each eval. This issue extends that with custom script execution.

Proposed Config

- name: claude-eval
  provider: claude-code
  workspace_template: ./templates/workspace/
  workspace_setup: ./scripts/setup.sh      # runs after copy, before eval
  workspace_teardown: ./scripts/cleanup.sh # runs after eval, before deletion

Use Cases

1. Install dependencies: npm install, pip install -r requirements.txt
2. Generate files: Create config files from templates
3. Database setup: Seed test data
4. Custom cleanup: Archive logs, save artifacts before deletion

Requirements

• Scripts can be any executable (shell, python, node, etc.)
• Scripts receive workspace path as argument or env var
• Setup script failure should abort the eval case
• Teardown script failure should warn but not fail the eval
• Timeout support for scripts
• Capture script output in eval results

Alternatives Considered

• CLAUDE.md instructions: Agent runs setup, but non-deterministic
• allagents integration: External tool, adds dependency
• Built-in scripts: Simpler but covers most cases

Related

• Depends on: PR feat: add workspace_template for isolated eval workspaces #180 (workspace_template feature)