storage: benchmark zstd vs snappy

WIth zstd now implemented, we'd like to know whether / when to recommend that to customers and whether we should consider making zstd the default compression.

Of note, we believe that zstd has better compression but higher CPU requirements. The goal of this benchmarking should be to run a few different types of workloads with different read/write ratios and row sizes to understand and approximate, at a high level:
- Compression improvements
- CPU usage differences

So that customers (and we) can make informed decisions on defaults and settings

Jira issue: CRDB-37668