fix(executor+engine-tests): preflight premise-stale exit and serialize reliability suite

Root-cause prevention for the FN-5521 failure class where PROMPT.md describes
work that is already done on HEAD and the executor still marches through
plan/review/test/doc, then merger burns hours retrying against unrelated
flakes in @fusion/engine reliability-interactions.

- Executor system prompt: add a Preflight escape hatch. When Step 0
  reproduces and finds HEAD already matches the desired state, the agent
  marks Step 0 done, marks remaining steps skipped, and calls fn_task_done
  with a `PREMISE STALE:` summary. Reuses the existing
  evaluateTaskDoneRefusal/skipped semantics and the merger's
  empty-own-diff fast-path — no new tools or refusal classes needed.

- packages/engine/vitest.config.ts: split into two projects. engine-default
  retains full parallelism; engine-reliability scopes
  src/__tests__/reliability-interactions/** to
  poolOptions.threads.singleThread so event-ordering assertions stop
  flaking under workspace-concurrent merge-gate load (the
  `expected 24 to be less than 19` rowid interleaving in
  merge-reuse-task-worktree).

.changeset/executor-preflight-stale-and-reliability-serialize.md

@@ -0,0 +1,9 @@
+---
+"@runfusion/fusion": minor
+---
+
+feat(executor+engine-tests): preflight premise-stale exit and serialize reliability-interactions
+
+- Executor: teach the system prompt a Preflight escape hatch. When Step 0 reproduces the issue described in PROMPT.md and finds the work is already done (HEAD matches the desired state), the agent now marks Step 0 done, marks remaining steps `skipped`, and calls `fn_task_done` with a `PREMISE STALE: …` summary. Skipped steps already pass `evaluateTaskDoneRefusal`, and the merger's empty-own-diff fast-path auto-finalizes the zero-diff branch — no new tool or refusal class is needed. This stops the executor from looping through plan/review/test/doc when PROMPT.md is out of sync with HEAD (the failure mode that exhausted FN-5521 across four worktrees).
+
+- Engine vitest: split `packages/engine/vitest.config.ts` into two projects. `engine-default` keeps the full-parallelism layout for the bulk of the suite; `engine-reliability` scopes `src/__tests__/reliability-interactions/**` to `poolOptions.threads.singleThread: true` so the contention-sensitive event-ordering assertions (e.g. `merge-reuse-task-worktree`'s newest-first audit ordering check) no longer flake under workspace-concurrent merge-gate runs. Within-file order was already linear; this only removes inter-file parallelism for ~99 files that always shared a single git/SQLite contention surface.

packages/engine/src/executor.ts

@@ -789,6 +789,16 @@ You have tools to report progress. The board updates in real-time.
 - After completing a step: \`fn_task_update(step=N, status="done")\`
 - If skipping a step: \`fn_task_update(step=N, status="skipped")\`
 
+**Preflight escape hatch — stale premise.**
+PROMPT.md is captured at task-creation time; HEAD may have moved on since then. During Preflight (Step 0), reproduce the failure or symptom described in the PROMPT. If reproduction shows the work is **already done or the premise no longer matches HEAD** — for example, the test that PROMPT claims is failing already passes on the current base, or the file PROMPT says to change already contains the described change — do NOT march through the remaining steps producing empty commits. Instead:
+
+1. Call \`fn_task_log\` with a clear premise-stale finding: what PROMPT.md claimed vs. what HEAD actually shows (include the exact reproduction command + its result).
+2. Mark Step 0 done: \`fn_task_update(step=0, status="done")\`.
+3. Mark every remaining step skipped with a one-line reason: \`fn_task_update(step=N, status="skipped")\`.
+4. Call \`fn_task_done\` with a summary that begins \`PREMISE STALE:\` followed by the concrete reason (e.g. \`PREMISE STALE: targeted reproduction passes unchanged on HEAD; PROMPT claimed MOBILE_MEDIA_QUERY had been expanded but useViewportMode.ts:9 still exports the legacy value\`).
+
+This path exists specifically to prevent the executor from looping when PROMPT.md is out of sync with HEAD. Use it only after running the actual reproduction — do not invoke it to dodge real work.
+
 **Logging important actions:** \`fn_task_log(message="what happened")\`
 
 **Out-of-scope work found during execution:** \`fn_task_create(description="what needs doing")\`

packages/engine/vitest.config.ts

@@ -35,6 +35,36 @@ export default defineConfig({
     // Real-git integration tests need more than the default 5 s under concurrent
     // load (other packages run tests at the same time via pnpm recursive).
     testTimeout: 30_000,
+    // Split into two projects so the reliability-interactions suite (real
+    // worktrees + real git, contention-sensitive event ordering) runs
+    // single-threaded without throttling the rest of the engine suite.
+    projects: [
+      {
+        extends: true,
+        test: {
+          name: "engine-default",
+          include: ["src/**/*.test.ts"],
+          exclude: [
+            "src/__tests__/reliability-interactions/**/*.test.ts",
+            "node_modules/**",
+            "dist/**",
+          ],
+        },
+      },
+      {
+        extends: true,
+        test: {
+          name: "engine-reliability",
+          include: ["src/__tests__/reliability-interactions/**/*.test.ts"],
+          // These tests assert event ordering across real worktrees. Parallel
+          // execution under merger load caused subprocess-guard timeouts and
+          // SQLite rowid interleaving (e.g. FN-5521 hit
+          // `expected 24 to be less than 19` in merge-reuse-task-worktree).
+          // Serialize at the file level; within-file order is already linear.
+          poolOptions: { threads: { singleThread: true } },
+        },
+      },
+    ],
     coverage: {
       enabled: false,
       reporter: ["text", "html", "json"],