Honor container settings in task takeover

[eshulman2]

Summary

• stop constructing a hard-coded default ContainerConfig in task-takeover execution
• apply the same fix to task-takeover qualitative review
• let ContainerRunner derive timeout, image, memory, and CPU limits from application settings
• add regression assertions that these workflows do not override runner defaults

Root cause

Both task-takeover call sites passed ContainerConfig(), whose timeout defaults to 1800 seconds. Because a config object was supplied, ContainerRunner.run() never called its settings-backed _default_config(), so CONTAINER_TIMEOUT=7200 was ignored and Podman received --timeout 1800.

Testing

• Ruff passed for all four changed files
• Focused unit suite: 23 passed