fix(agent-loop): complete direct answers without a fake tool retry

[imMamdouhaboammar]

Problem

When the model returns a final text response with no tool calls, agent_runner_loop() creates the internal pseudo-call no_tool. GenericAgentHandler.turn_end_callback() explicitly recognizes that pseudo-call as a direct answer, but BaseHandler.dispatch() did not. It fell through to the unknown-tool branch, producing 未知工具 no_tool and forcing another turn.

With max_turns=1, a valid direct answer therefore ended as MAX_TURNS_EXCEEDED instead of CURRENT_TASK_DONE.

Fix

Teach BaseHandler.dispatch() the existing internal no_tool sentinel and return a normal StepOutcome with no next prompt. The loop then follows its existing completion path. Done hooks remain supported because CURRENT_TASK_DONE still passes through the existing done-hook logic.

Verification

TDD was performed on a separate validation branch:

1. Regression-only GitHub Actions run 31165930923 failed on current main: the direct answer returned MAX_TURNS_EXCEEDED instead of CURRENT_TASK_DONE
2. The one-line dispatch fix passed run 31165988879: compile, focused regression, and git diff --check
3. The final contribution branch was rebuilt directly from current main as one clean commit containing only the production change and focused regression test

Scope

• 1 core production file modified
• 1 focused regression-test file added
• 1 clean commit on top of current main
• no dependency or configuration changes