`load data` in cluster mode may cause out-of-order insertion, leading poor performance for long window

**Bug Description**
if `spark` is configured with multiple threads, say `local[*]` or `local[32]`, data loading will run in parallel, causing out-of-order loading.

It is ok for normal data insertion. But `long window optimization` requires that data is loaded mostly in order.

**Potential solution**
- remove the `in order loading` requirement for `long window`, by deleting the old entry and inserting the new one in the pre-aggr table.