Expose shouldStopAfterTurn on Agent

[xlr8harder]

Summary

• expose the existing loop-level shouldStopAfterTurn callback through public AgentOptions
• retain the callback on Agent and forward it into AgentLoopConfig
• cover the public behavior with a tool turn that records its result, stops, and avoids a second provider call

Why

AgentLoopConfig already supports stopping at a completed turn boundary, but callers of the public Agent API cannot configure that hook. They currently have to subclass Agent and override the private _create_loop_config method.

This keeps the public wrapper aligned with the loop API while preserving the existing ordering: the callback runs after tool execution and prepareNextTurn, and before another provider turn.

Testing

• uvx ruff check packages/harn_agent/src/harn_agent/agent.py tests/test_agent/test_agent_api.py
• uv run --all-packages pytest tests/test_agent -q — 83 passed
• deterministic repository suite — 1377 passed, 2 unrelated environment-sensitive tests deselected
• uv build --package harn-agent