Add optional per-task agent trajectory export

[Sehlani042]

Summary

• add an optional --trajectory-output-dir flag for saving per-task agent trajectories
• write trajectory text files grouped by model name without changing default benchmark behavior
• include task metadata plus the execution summary used for judge evaluation
• document the CLI option and add lightweight unittest coverage for path sanitization, file writing, and runner integration

Tests

• .venv/bin/python -m unittest discover -s tests
• .venv/bin/python -m compileall benchmark tests/test_trajectory_writer.py tests/test_runner_trajectory_export.py
• git diff --check

Addresses #18.
Addresses #9.