fix: increase default max iterations for agents to improve performanc…

[pikann]

Summary

• Increase default and maximum iteration limits for agents — default max_iterations raised from 50 → 500, and the upper cap from 200 → 500, both in the Go service and the DB migration.
• Fix 400 errors from sandbox on conversation lookup — removed conversation_id from the Conversation constructor. The sandbox server assigns its own internal UUID; passing our application ID caused GET /api/conversations/{id} to return 400 on all subsequent calls. Event callbacks already persist to our DB using trigger.conversation_id directly, so the mapping stays correct.
• Fix lingering WebSocket thread after sandbox shutdown — added conversation.close() in the executor finally block. Without this, the SDK's WebSocket client thread outlived the sandbox container; when Docker recycled the container IP, the thread would send requests for the old conversation ID to the new container and receive 400s.

Changes

File	Change
services/api/internal/service/agent/agent_service.go	maxIterationsLimit 200 → 500; default 50 → 500 (both CreateAgent and UpdateAgent)
services/api/migrations/000008_add_ai_agents.sql	max_iterations column default 50 → 500
services/ai-agent/src/agent/executor.py	Remove conversation_id arg; add conversation.close() in finally; remove unused import uuid

Test plan

• Create a new agent and confirm max_iterations defaults to 500
• Trigger an agent run and verify no 400 errors appear in logs for conversation lookup
• Stop a sandbox mid-run and confirm no lingering WebSocket thread errors in subsequent runs

🤖 Generated with Claude Code