Generate a skewed workload at shard granularity in mako

Mako's `-z` option can generate a skewed workload based on Zipf distribution, but this is at the record level. It would be useful to skew workloads over shards, rather than records, so some are busier than others.

A new option `-zs` could divide the record count by the default shard size. While running, `zipfian_next()` could select a shard from that set. Then a random record could be selected from that shard.