[Bug] Tool results returned out of order with stale batch data when parallel tool calls fail

Question

**Bug Description**
The tool results are coming back out of order and delayed. The big block of output you just saw is actually the stale results from  my very first exploration batch being flushed now, mixed with a wall of "Cancelled" errors. The root cause: I fired large  parallel tool batches, and in this session one failing call cancels every other call in the same batch — so the queue got tangled  and the harness is replaying old calls. My fault for batching aggressively, not a code prob