refactor(openapi): co-locate per-mode run response models in _models.py

Prepares for the api-3 unified /run route which imports all three
response shapes from one location. The per-mode files (chat_messages,
completion_messages, workflow_run) still define their own copies inline
until Task 4 deletes them — no collision since neither location is
imported anywhere else.

api/controllers/openapi/_models.py

@@ -4,7 +4,7 @@ from __future__ import annotations
 
 from typing import Any
 
-from pydantic import BaseModel
+from pydantic import BaseModel, Field
 
 # Server-side cap on `limit` query param for any /openapi/v1/* list endpoint.
 # Sibling endpoints (`/apps`, `/account/sessions`, future routes) all clamp to
@@ -67,3 +67,46 @@ class AppDescribeResponse(BaseModel):
     info: AppDescribeInfo | None = None
     parameters: dict[str, Any] | None = None
     input_schema: dict[str, Any] | None = None
+
+
+class ChatMessageResponse(BaseModel):
+    event: str
+    task_id: str
+    id: str
+    message_id: str
+    conversation_id: str
+    mode: str
+    answer: str
+    metadata: MessageMetadata = Field(default_factory=MessageMetadata)
+    created_at: int
+
+
+class CompletionMessageResponse(BaseModel):
+    event: str
+    task_id: str
+    id: str
+    message_id: str
+    mode: str
+    answer: str
+    metadata: MessageMetadata = Field(default_factory=MessageMetadata)
+    created_at: int
+
+
+class WorkflowRunData(BaseModel):
+    id: str
+    workflow_id: str
+    status: str
+    outputs: dict[str, Any] = {}
+    error: str | None = None
+    elapsed_time: float | None = None
+    total_tokens: int | None = None
+    total_steps: int | None = None
+    created_at: int | None = None
+    finished_at: int | None = None
+
+
+class WorkflowRunResponse(BaseModel):
+    workflow_run_id: str
+    task_id: str
+    mode: str = "workflow"  # echoed for CLI per-mode rendering — see endpoints.md L154
+    data: WorkflowRunData

api/tests/unit_tests/controllers/openapi/test_pagination_envelope.py

@@ -100,3 +100,19 @@ def test_app_describe_response_nests_info_and_parameters():
     dumped = obj.model_dump(mode="json")
     assert dumped["info"]["service_api_enabled"] is True
     assert dumped["parameters"]["opening_statement"] is None
+
+
+def test_response_models_dump_per_mode():
+    from controllers.openapi._models import (
+        ChatMessageResponse, CompletionMessageResponse, WorkflowRunResponse, WorkflowRunData,
+    )
+    chat = ChatMessageResponse(
+        event="message", task_id="t1", id="m1", message_id="m1",
+        conversation_id="c1", mode="chat", answer="hi", created_at=0,
+    )
+    assert chat.model_dump(mode="json")["mode"] == "chat"
+    wf = WorkflowRunResponse(
+        workflow_run_id="r1", task_id="t1",
+        data=WorkflowRunData(id="r1", workflow_id="w1", status="succeeded"),
+    )
+    assert wf.model_dump(mode="json")["data"]["status"] == "succeeded"