Add support for overridding the number of records in the workload

[daverigby]

To allow users to simulate workloads which are similar to existing (but not necessarily identical), it would be valuable to be able to override the number of records which will be populated on the command - say via a --records=N option.

This should be passed to the Workload being instantiated, which should then modify it's behaviour to use the given number of records.

For ParquetWorkloads (i.e. the only type we have so far), it should only be valid to set the number of workloads to less than or equal to how many exist in the parquet files. Additionally, if this option is specified then it will become necessary to recalculate the ground-truth nearest neighbours for the query set (as per ParquetSubSetWorkload).

(Note: similar to #184)