docs(run-engine): correct the reasoning on the acknowledged-run guard

The guard that stops a returned message being written back when its payload is
gone was described as cleaning up an orphaned list entry. It is load-bearing:
an acknowledgement landing between the candidate read and the script leaves a
claimable worker queue entry whose run has already completed, and without the
guard the sweep would put that completed run back on the pending queue.

Also aligns two test doubles with the renamed result field.

Author: matt-aitken

apps/webapp/test/pauseSweepWiring.test.ts

@@ -24,7 +24,7 @@ vi.mock("~/v3/runQueue.server", () => ({
   }),
   returnUnclaimedMessagesToQueue: vi.fn(async () => ({
     returned: 0,
-    skipped: 0,
+    skippedLastPass: 0,
     errors: 0,
     passes: 1,
   })),

apps/webapp/test/sweepUnclaimedRuns.test.ts

@@ -24,7 +24,7 @@ describe("sweepUnclaimedRuns", () => {
 
     returnUnclaimedMessagesToQueue.mockResolvedValueOnce({
       returned: 2,
-      skipped: 0,
+      skippedLastPass: 0,
       errors: 0,
       passes: 1,
     });

internal-packages/run-engine/src/run-queue/index.ts

@@ -4846,7 +4846,8 @@ if claimed == 0 then
   return 0
 end
 
--- An orphaned list entry (payload already acknowledged): the LREM above cleaned it up.
+-- Acknowledged between the candidate read and this script. The LREM above removed the
+-- stale entry; the completed run must not be written back onto the queue.
 if redis.call('EXISTS', messageKey) == 0 then
   return 0
 end
@@ -4927,7 +4928,8 @@ if claimed == 0 then
   return 0
 end
 
--- An orphaned list entry (payload already acknowledged): the LREM above cleaned it up.
+-- Acknowledged between the candidate read and this script. The LREM above removed the
+-- stale entry; the completed run must not be written back onto the queue.
 if redis.call('EXISTS', messageKey) == 0 then
   return 0
 end