M8：天工阁、叩心与辰正——受托任务后台并行、叩心四约束、时态感知投影

- 天工阁域 domains/tiangong：backstage_tasks（委托包+验收标准+结果证据）；有限分解（parent 链）与受控并行（各自后台 Run）；子任务全终态级联父任务进入待验收；按证据 accept/reject
- kernel.start_backstage_run（kind=background_task）：loop 后台分支不占前台席位、与前台对话并行；断言放宽容纳无人物功能任务；结果只经回调回呈任务——无身份写入路径
- 叩心域 domains/kouxin：knocking_candidates（event/time_anchor/manual 触发 + 草稿 + 裁决快照）；四约束裁决（总开关→静默窗口【本地时区判定】→24h 频率→预算），dispatch 未过 429 且快照留痕；候选默认不自动发送；expire_stale 过期推进不触碰缘契
- 辰正 domains/zhengchen：temporal_context() 复用 time_sense 计算核（快照/农历/季节），产出【时态感知】投影段接入镜树生成 prompt（客观时间语境，不附情绪臆测——守一纪律）
- 退役：modules/goal_orchestrator|planner|perception|big_cycle、application/planning、services/governance|goal_persist、candidates 表链（imports 候选抽取 API 面）、goals/plans/agent_roster/perception_channels/big_cycles/旧候选表（迁移 009）；前端 admin goals/plans/agents/perception/world 页与六组件；governance 只留 GenerationPreferences；admin 治理面收敛为司命+百工两枢；pivots 健康以域表计数重建
- 验收（test_m8_tiangong 五用例）：后台任务无身份写入（记忆/魂籍/缘契零新增）；后台与前台并行；分解级联；叩心四约束逐项；不回应不降低关系（缘契维度/状态/版本零变化）
- 四套件：后端 345、前端 203、BFF 4、神识 5 包全绿（2 个预存环境失败不变）；迁移 009 净库干净

backend/app/schemas/candidates.py

@@ -1,50 +0,0 @@
-"""候选 API schema（api-spec §8）。"""
-
-from pydantic import BaseModel, Field
-
-
-class ProfileCandidateResponse(BaseModel):
-    id: str = Field(description="候选 ID，前缀 profile_candidate_。")
-    character_id: str = Field(description="所属人物 ID。")
-    import_job_id: str | None = Field(default=None, description="来源导入任务 ID。")
-    field: str = Field(description="档案字段名。")
-    value: str = Field(description="候选值。")
-    confidence: float = Field(description="置信度。")
-    evidence_ids: list[str] = Field(default_factory=list, description="证据 import_item ID。")
-    status: str = Field(description="审核状态，契约字典 §10.6。")
-    reason: str | None = Field(default=None)
-    reviewed_at: str | None = Field(default=None)
-    created_at: str = Field(description="创建时间，UTC ISO 8601 字符串。")
-
-
-class MemoryCandidateResponse(BaseModel):
-    id: str = Field(description="候选 ID，前缀 memory_candidate_。")
-    character_id: str = Field(description="所属人物 ID。")
-    import_job_id: str | None = Field(default=None, description="来源导入任务 ID。")
-    cycle_id: str | None = Field(default=None, description="来源大周天循环 ID。")
-    type: str = Field(description="记忆类型。")
-    title: str = Field(description="记忆标题。")
-    content: str = Field(description="记忆内容。")
-    confidence: float = Field(description="置信度。")
-    importance: float = Field(description="重要度。")
-    sensitivity: str = Field(description="敏感等级。")
-    evidence_ids: list[str] = Field(default_factory=list, description="证据 import_item ID。")
-    status: str = Field(description="审核状态，契约字典 §10.6。")
-    reason: str | None = Field(default=None)
-    reviewed_at: str | None = Field(default=None)
-    created_at: str = Field(description="创建时间，UTC ISO 8601 字符串。")
-
-
-class CandidateReviewDecision(BaseModel):
-    candidate_id: str = Field(description="候选 ID。")
-    decision: str = Field(description="审核决定：approve / reject。")
-
-
-class BatchReviewRequest(BaseModel):
-    decisions: list[CandidateReviewDecision] = Field(min_length=1, description="批量审核决定。")
-
-
-class BatchReviewResult(BaseModel):
-    applied: int = Field(description="已应用（approve）数量。")
-    rejected: int = Field(description="已驳回数量。")
-    skipped: int = Field(description="因状态非 pending 跳过的数量。")