feat(agent-scan): add jailbreak detection support (#303)

[boy-hack]

Summary

Closes #303

This PR adds optional jailbreak detection to the Agent Scan workflow, and implements the standalone ModelJailbreak task type.

Changes

common/websocket/api.go

• Add jailbreak bool field to AgentScanTaskRequest
• Pass the flag through to task params
• Update Swagger docs with agent_scan example including jailbreak field
• Update task type description to include Agent Scan (was missing)

common/agent/agent_task.go

• Add Jailbreak bool to AgentScanParams
• When jailbreak=true, add an extra "Jailbreak Detection" step to the plan
• After agent scan completes successfully, invoke AIG-PromptSecurity/cli_run.py with Custom:prompt scenario
• Jailbreak step failure is non-fatal — agent scan results remain valid

common/agent/jailbreak_task.go (new)

• Implement ModelJailbreak struct with GetName() and Execute()
• Standalone jailbreak task: accepts model.{model,token,base_url} + prompt
• Invokes AIG-PromptSecurity/cli_run.py with Raw technique, serial choice

cmd/agent/main.go

• Register ModelJailbreak handler alongside existing task handlers

Notes

⚠️ This PR has a minor overlap with PR #330 (api.go). If #330 merges first, this branch will need a git rebase origin/main to resolve the conflict (trivial — different struct fields).

Test

# Agent scan with jailbreak detection enabled
curl -X POST http://localhost:8080/api/v1/app/taskapi/tasks \
  -H "Content-Type: application/json" \
  -d '{
    "type": "agent_scan",
    "content": {
      "agent_id": "my-agent",
      "eval_model": {"model": "gpt-4", "token": "sk-xxx", "base_url": "https://api.openai.com/v1"},
      "language": "en",
      "jailbreak": true
    }
  }'