Recheck feed-in loop stall fix in doOneFeedInTask

[Copilot]

The original feed-in loop silently skipped out-of-range dataIndex without advancing portRounds[i], making it possible for a task to stall indefinitely if stride/port/data-length were inconsistent.

What was verified

• Upfront validation — validateTaskLayout (called in FeedIn before enqueueing) rejects any inconsistent stride/portCount/dataLen combination immediately with a descriptive panic. No stalling task ever enters the queue.
• Runtime defensive check — the original silent continue was replaced with an explicit panic carrying full diagnostics (idx, len, stride, port, round):

if dataIndex < 0 || dataIndex >= len(task.data) {
    panic(fmt.Sprintf(
        "feed task index out of range: idx=%d len=%d stride=%d port=%d round=%d",
        dataIndex, len(task.data), task.stride, i, task.portRounds[i],
    ))
}

Because validateTaskLayout mathematically guarantees portRounds[i]*stride + i < dataLen for all valid (i, round) pairs, the runtime check is unreachable in correct usage — it is a safety net against future regressions, not an active code path.

No further changes are required; the current implementation fully addresses the stall concern.

---

🔒 GitHub Advanced Security automatically protects Copilot coding agent pull requests. You can protect all pull requests by enabling Advanced Security for your repositories. Learn more about Advanced Security.