Add non-uniform query access pattern(s)

[daverigby]

Currently the queries are issued in sequential order in the same order as they occur in the queries parquet dataset.

This doesn't represent many real-world workloads, which are less uniform and often exhibit a Zipfian-like distribution.

Add support for specifying a non-uniform query distribution.

(Note this also relates to concurrent access patterns by multiple independent clients - in a typical real-world scenario one would observe some non-zero, but non-100% correlation between different clients' access patterns - i.e. the most popular vectors are likely to be accessed by a large number of clients, but the least popular may only be accessed by a single client during a benchmark run.