AtomicLong.updateAndGet()超级费CPU，并发一多就100%


原来：
```
// spin get next available cursor
long currentCursor = cursor.get();
long nextCursor = cursor.updateAndGet(old -> old == tail.get() ? old : old + 1);

// trigger padding in an async-mode if reach the threshold
long currentTail = tail.get();
```

改成下面这样是否可以？？？
```
// spin get next available cursor
long currentCursor = cursor.get();
long currentTail = tail.get();

long nextCursor = currentCursor;
if (currentCursor != currentTail) {
nextCursor = cursor.incrementAndGet();
}
```

Start from the cursor updateAndGet() path shown in the issue and examine its behavior as concurrency increases. Compare the proposed get/increment sequence with the existing atomic update under contention, checking cursor and tail correctness as well as CPU usage. Done means the chosen approach is validated for concurrent behavior and the reported CPU problem is addressed.

Written by the indexing model from the issue text.