[SPARK-18990][SQL] make DatasetBenchmark fairer for Dataset

## What changes were proposed in this pull request?

Currently `DatasetBenchmark` use `case class Data(l: Long, s: String)` as the record type of `RDD` and `Dataset`, which introduce serialization overhead only to `Dataset` and is unfair.

This PR use `Long` as the record type, to be fairer for `Dataset`

## How was this patch tested?

existing tests

Author: Wenchen Fan <wenchen@databricks.com>

Closes #16391 from cloud-fan/benchmark.