Preserve Stirrup code-execution workspaces for scenario-suite runs

[ChathurangiShyalika]

This PR adds opt-in preservation for Stirrup code-execution workspaces so benchmark artifacts can be inspected after each scenario run.

Changes:

• Adds --workspace-dir and --preserve-workspace to stirrup-agent.
• Passes the workspace path into Stirrup’s Docker/local code backend.
• Copies final code_exec files back into the requested workspace before Stirrup cleanup.
• Adds scenario-suite flags: --stirrup-workspace-root and --preserve-workspaces

Keeps per-scenario workspaces under:
<root>/stirrup_agent/<model-dir>/stirrup_agent_<scenario_id>/

Example:

uv run python -m benchmark.scenario_suite_runner \
  --scenario-ids benchmarks/scenario_suite/scenarios.txt \
  --scenario-root /.../scenarios_data \
  --agent_name stirrup_agent \
  --model-id tokenrouter/z-ai/glm-5.2 \
  --trajectory-root /tmp/assetopsbench-trajectories \
  --reports-root /tmp/assetopsbench-reports \
  --stirrup-workspace-root /tmp/assetopsbench-stirrup-workspaces \
  --preserve-workspaces \
  --continue-on-error

Validation:

• Added focused tests for Stirrup workspace preservation.
• Confirmed CLI help exposes the new flags.
• Verified focused tests pass.