M4b：ai_provider 删除手术①-⑥——神识全面接管模型出口

- ① 后端删除 app/ai_provider/（38 文件，含 anthropic/openai 格式、billing、audio、embedding）、services/ai_provider_admin.py、core/config/model_config.py、tests/ai_provider/（21 文件）
- ② 连带退役 application/agent/（旧 Agent 循环/steer/follow-up/确认注册表）、application/skill_executor.py、app/skills/（python skills 孤儿）、endpoints/audio.py、application/imports.py（ProviderManager 消费方）；settings 端点收敛为生成偏好
- ③ runs.py 修剪：AgentRunRegistry/steer/follow-up/confirmations 端点退役，保留 list/get/events/cancel/stream；safety 确认改为直接落库裁决（pending→approved/denied）
- ④ 前端删除：models 页、ModelGatewayView、gateway.ts、GatewayPanel（玄关）、ModelGatewayPanel/ModelStrategyForm、audio.ts/useVoiceInput、网关类型与 mock；设置页收敛为外观/语言/通知
- ⑤ 配置收敛：config/models|providers|pricing.example.yaml 归档 shenshi/config/；.env.example 删 provider keys/MODEL_ENDPOINT_POOL/MODEL_CONFIG_PATH，收敛 SHENSHI_*
- ⑥ 验收：rg 审计灵台 Provider SDK import 与 ai_provider 引用归零；后端 528 passed（2 个预存失败经 HEAD 复跑验证与本手术无关）、前端 type-check+201 passed、神识 go vet+test、BFF 4 passed

backend/tests/api/test_model_gateway_endpoints.py

@@ -1,78 +0,0 @@
-"""模型网关 endpoint 覆盖 API 测试（api-spec §4.4，DB 优先于 .env/yaml）。"""
-
-from collections.abc import AsyncGenerator
-
-import pytest
-from httpx import ASGITransport, AsyncClient
-
-from app.main import app
-
-
-@pytest.fixture
-async def client() -> AsyncGenerator[AsyncClient, None]:
-    transport = ASGITransport(app=app)
-    async with AsyncClient(transport=transport, base_url="http://test") as http_client:
-        yield http_client
-
-
-def _first_endpoint(gateway: dict) -> tuple[str, str]:
-    model = gateway["models"][0]
-    return model["name"], model["endpoints"][0]["name"]
-
-
-@pytest.mark.asyncio
-async def test_list_models_has_endpoint_override_fields(client: AsyncClient) -> None:
-    resp = await client.get("/api/v1/settings/models")
-    assert resp.status_code == 200
-    ep = resp.json()["models"][0]["endpoints"][0]
-    assert "priority" in ep
-    assert "is_overridden" in ep
-    assert "has_url_override" in ep
-    assert "has_key_override" in ep
-    assert ep["is_overridden"] is False
-
-
-@pytest.mark.asyncio
-async def test_patch_endpoint_status_override(client: AsyncClient) -> None:
-    gateway = (await client.get("/api/v1/settings/models")).json()
-    model, endpoint = _first_endpoint(gateway)
-
-    resp = await client.patch(
-        f"/api/v1/settings/models/{model}/endpoints/{endpoint}",
-        json={"status": "inactive"},
-    )
-    assert resp.status_code == 200
-    body = resp.json()
-    assert body["is_overridden"] is True
-    assert body["status"] == "inactive"
-
-    # GET 反映覆盖
-    gateway2 = (await client.get("/api/v1/settings/models")).json()
-    eps = [m for m in gateway2["models"] if m["name"] == model][0]["endpoints"]
-    ep = [e for e in eps if e["name"] == endpoint][0]
-    assert ep["is_overridden"] is True
-    assert ep["status"] == "inactive"
-
-
-@pytest.mark.asyncio
-async def test_patch_endpoint_key_override_redacted(client: AsyncClient) -> None:
-    gateway = (await client.get("/api/v1/settings/models")).json()
-    model, endpoint = _first_endpoint(gateway)
-
-    resp = await client.patch(
-        f"/api/v1/settings/models/{model}/endpoints/{endpoint}",
-        json={"api_key": "sk-test-secret-value"},
-    )
-    assert resp.status_code == 200
-    assert resp.json()["has_key_override"] is True
-    # 脱敏：响应体不含 api_key 原值
-    assert "sk-test-secret-value" not in resp.text
-
-
-@pytest.mark.asyncio
-async def test_patch_endpoint_not_found(client: AsyncClient) -> None:
-    resp = await client.patch(
-        "/api/v1/settings/models/no_such_model/endpoints/primary",
-        json={"status": "inactive"},
-    )
-    assert resp.status_code == 404