Improve agent investigation with shared-profile peer evidence

[eirenik0]

Summary

• Add shared-profile peer evidence to incident investigations so healthy sibling containers weaken account-wide suspicion.
• Extend investigation context and fallback analysis with explicit healthy/unhealthy peer classification.
• Update operator guidance to prefer service-specific remediation when one container is unhealthy but peers with the same profile are healthy.
• Add coverage for the new peer-evidence behavior and the resulting investigation summaries/action plans.

Testing

• make fmt
• make lint
• uv run pytest -q tests/test_agent_runtime.py tests/test_agent_cli.py tests/test_server_monitor.py tests/test_fleet_cli.py
• uv run pytest -q tests/test_fleet_state_manager.py