feat: add Terminal-Bench environment

[ZixuanZZhang]

Summary

Adds environment support for Terminal-Bench 2.0, a benchmark for evaluating AI agents on real-world terminal tasks. Each task runs in an isolated Docker container with the agent interacting via shell commands. The same environment can be used for Terminal-Bench benchmark.

Changes: new environment: TerminalBenchEnv

• Docker-based environment using Harbor for container management
• Provides execute_command tool for shell interaction within containers
• Automatic container lifecycle management (build, start, cleanup)
• Binary reward function based on test script execution

Test results:

All existing unit tests pass (102 passed in 4.15s)

Additional dependencies:

Require Harbor for Docker environment management.

[Lawhy]

LGTM!