feat(benchmark): add scenario suite runner

[ChathurangiShyalika]

Summary

Adds a generic benchmark runner for scenario-suite execution. The runner reads a simple scenario-id list, runs each scenario sequentially, resets and reloads CouchDB uniformly for every scenario, saves trajectories, and invokes the existing evaluator to generate per-scenario and aggregate reports.

Changes

• Added benchmarks/scenario_suite/scenarios.txt as the scenario-id list
• Added benchmarks/scenario_suite/scenarios.sample.txt as a public example file
• Added benchmarks/scenario_suite/README.md
• Added src/benchmark/scenario_suite_runner.py
• Added src/benchmark/tests/test_scenario_suite_runner.py
• Updated direct LLM agent files to support the benchmark flow
• Updated evaluation report aggregation for static JSON results
• Updated related tests

Behavior

• Reads scenario IDs from benchmarks/scenario_suite/scenarios.txt
• Uses SCENARIOS_DATA_DIR from the provided scenario root
• Resets and reloads CouchDB for every scenario so the flow stays uniform
• Runs the selected agent method sequentially
• Saves trajectories under AGENT_TRAJECTORY_DIR
• Runs the evaluator with --scorer-default static_json
• Writes per-scenario reports and an aggregate report

Notes

• The public sample file is intentionally generic and does not expose private benchmark wording.
• Scenario data remains external and is not committed in full.

[DhavalRepo18]

I cross-check every file. We can get started.