Add sandbox work-folder persistence and shared lifecycle

Introduce company-scoped task, agent, user and project files, recoverable trash, object-backed repository checkpoints, and a shared 180-second lifecycle for sandbox execution. Wire file browsing and save status into the owner surfaces. This is an implementation checkpoint; staging acceptance and the complete merge gate remain outstanding.

server/src/services/work-folder-checkpointer.ts

@@ -0,0 +1,38 @@
+import { WORK_FOLDER_SYNC_INTERVAL_MS } from "@paperclipai/shared";
+
+/** One scheduler for both execution generations. A tick never queues a backlog. */
+export function startWorkFolderCheckpointer(input: {
+  checkpoint(): Promise<void>;
+  onError(error: unknown): Promise<void>;
+}) {
+  let active: Promise<void> | null = null;
+  let stopped = false;
+  function checkpoint() {
+    const pending = (async () => {
+      try { await input.checkpoint(); }
+      catch (error) { await input.onError(error); throw error; }
+    })();
+    active = pending;
+    void pending.finally(() => { if (active === pending) active = null; }).catch(() => {});
+    return pending;
+  }
+  const timer = setInterval(() => {
+    if (!stopped && !active) void checkpoint().catch(() => {});
+  }, WORK_FOLDER_SYNC_INTERVAL_MS);
+  timer.unref();
+  // Explicit flushes serialize too. They must checkpoint once more after an
+  // in-flight tick, since the agent may have edited during that tick.
+  let flushTail: Promise<void> = Promise.resolve();
+  function flush() {
+    const pending = flushTail.catch(() => {}).then(async () => {
+      await active?.catch(() => {});
+      await checkpoint();
+    });
+    flushTail = pending;
+    return pending;
+  }
+  return {
+    flush,
+    async stop() { stopped = true; clearInterval(timer); await flush(); },
+  };
+}