feat: CHV gap fill — worker breakpoint wiring, pool reconcile, per-VM host keys, egress validation, guest build script

Stack: #216 -> #221 -> this. Worker breakpoint branch now calls any
executor providing goal_resume_breakpoint_task (Local keeps its
original path, gateway tasks still loud-fail); WarmPool
reconcile_orphans is called on CHV worker startup and close with
audit-on-failure; per-VM SSH host keys are provisioned before VMM
start via cloud-init NoCloud seed (keyscan kept only as weak
fallback, CHV_HOST_KEY_MODE=provisioned|keyscan); CHV_NET_JSON is
validated in preflight and explicit direct egress is audited;
chv-guest/build.sh is now real steps with --dry-run and fails fast
without /dev/kvm. Tests: new test_chv_gaps.py (18), cloud full
1514 OK (skipped=4). No real VM verification: no KVM host.

cloud/worker.py

@@ -120,6 +120,7 @@ Docker 下 repos_dir 必须放在 extra-mounts 已挂载的目录下，
 
 import argparse
 import hashlib
+import inspect
 import json
 import os
 from datetime import datetime, timezone
@@ -332,6 +333,24 @@ def _merge_env(base, override):
     return merged
 
 
+def _call_goal_resume_breakpoint(executor, task, model_provider=None):
+    """通用断点调用：执行器提供 goal_resume_breakpoint_task 时调用它。
+
+    Local 仍走原路径（调用方直接传 model_provider，不经过这里，
+    以免改生产语义）；这里只用于非 Local 执行器：签名接受
+    model_provider 时传它（CHV/Docker 均接受），否则只传 task
+    （旧假执行器单参）。网关 key 绝不经这里传值：worker 侧
+    gateway_env 只按名注入，见 _run_attempt 的网关守卫。"""
+    method = getattr(executor, "goal_resume_breakpoint_task")
+    try:
+        params = inspect.signature(method).parameters
+    except (TypeError, ValueError):
+        return method(task)
+    if "model_provider" in params:
+        return method(task, model_provider=model_provider)
+    return method(task)
+
+
 def _resolve_model(model=None, config_path=None):
     """模型名：显式参数 > host 配置文件里的 model > host 默认值。"""
     if model:
@@ -525,6 +544,19 @@ class Worker:
                 self.executor.audit_callback = _chv_audit
             except Exception:
                 pass
+        # CHV 补缺口轨：WarmPool 显式接线。池构造本身不自动
+        # 清理（单活池假设，见 pool.py），worker 启动是显式
+        # 入口，在这里建池并调 reconcile_orphans()；关闭时
+        # close() 再调一次。调用失败不炸 worker 构造/关闭，
+        # 但要进审计（_audit 本身 best-effort）。
+        self._warm_pool = None
+        if isinstance(self.executor, CloudHypervisorExecutor):
+            try:
+                from cloud.chv.pool import WarmPool as _WarmPool
+                self._warm_pool = _WarmPool(self.executor)
+                self._reconcile_warm_pool("startup")
+            except Exception:
+                pass
         self._cost_store = cost_store
         self._cancel_requested = set()
         # 真取消用：process_factory 可注入（单测用它返回一个假长跑
@@ -759,8 +791,49 @@ class Worker:
             self._cost_store = CostStore(self.db_path)
         return self._cost_store
 
+    def _reconcile_warm_pool(self, when):
+        """显式调 WarmPool.reconcile_orphans()，返回已清理 VM 名。
+
+        仅 CHV worker 且池已建时有效；池未建时若执行器是 CHV
+        则现建现调（供测试与手动清理用）。调用失败不炸主流程，
+        但进审计（失败本身也审计，不静默丢弃）。orphan 条目
+        绝不发给任务，reconcile 只做清理。"""
+        pool = getattr(self, "_warm_pool", None)
+        if pool is None:
+            executor = getattr(self, "executor", None)
+            if not isinstance(executor, CloudHypervisorExecutor):
+                return []
+            try:
+                from cloud.chv.pool import WarmPool as _WarmPool
+                pool = _WarmPool(executor)
+                self._warm_pool = pool
+            except Exception:
+                return []
+        try:
+            cleaned = pool.reconcile_orphans() or []
+        except Exception as exc:
+            try:
+                self._audit(None, "chv_warm_pool_reconcile_failed",
+                            "when=%s error=%s" % (when, exc))
+            except Exception:
+                pass
+            return []
+        if cleaned:
+            try:
+                self._audit(None, "chv_warm_pool_reconcile",
+                            "when=%s cleaned=%s" % (when, ",".join(cleaned)))
+            except Exception:
+                pass
+        return cleaned
+
     def close(self):
-        """关闭惰性建的成本存储连接（注入的 cost_store 由调用方管）。"""
+        """关闭惰性建的成本存储连接（注入的 cost_store 由调用方管）。
+        CHV worker 关闭时先显式调一次 warm-pool reconcile，
+        失败不炸关闭流程。"""
+        try:
+            self._reconcile_warm_pool("shutdown")
+        except Exception:
+            pass
         if self._cost_store is not None:
             try:
                 self._cost_store._conn.close()
@@ -4712,6 +4785,18 @@ class Worker:
                                 self.executor, LocalExecutor):
                             result = self.executor.goal_resume_breakpoint_task(
                                 task, model_provider=gateway_provider)
+                        elif task.get("_goal_resume_breakpoint") and hasattr(
+                                self.executor,
+                                "goal_resume_breakpoint_task"):
+                            # CHV 补缺口轨：任何提供断点方法的执行器
+                            # 都可走断点（CHV 的 goal_resume_breakpoint_task
+                            # 已接线；Docker 走上面 is_docker 分支，不到
+                            # 这里）。网关任务在进入这里之前已被
+                            # “网关不支持该执行器”守卫 loud fail，
+                            # 故这里不放宽网关规则。
+                            result = _call_goal_resume_breakpoint(
+                                self.executor, task,
+                                model_provider=gateway_provider)
                         elif task.get("_goal_resume_breakpoint"):
                             # 假执行器/CloudExecutor 没有断点方法时
                             # 绝不回落成 goal（会撞已存在）或