[SPARK-15724] Add benchmarks for performance over wide schemas

## What changes were proposed in this pull request?

This adds microbenchmarks for tracking performance of queries over very wide or deeply nested DataFrames. It seems performance degrades when DataFrames get thousands of columns wide or hundreds of fields deep.

## How was this patch tested?

Current results included.

cc rxin JoshRosen

Author: Eric Liang <ekl@databricks.com>

Closes #13456 from ericl/sc-3468.