[Bug][SparkLoad] Spark load will create rowset with incorrect rowset type

**Describe the bug**
1. create a table with segment v2 format
2. load data with spark load
3. the rowset with version [2-2] is with storage format SegmentV1, which is expected to be V2.

**Why**

the push handler only set the rowset writer's rowset type by the config of `default_rowset_type` of BE,
not checking the `prefer_rowset_type` of the tablet.