new workflow: subsample_by_metadata_with_focal

[dpark01]

This PR adds a new WDL workflow, subsample_by_metadata_with_focal, which provides a functionality I've been doing manually/offline until recently. This subsets an externally provided dataset of sequences to a more manageable sample size with even representation across metadata-defined categories (e.g. geographic). It does so by splitting the external dataset into two categories: one termed focal and one termed global, and applying different sample count limits (and bin resolutions) to each, before remerging. Current defaults are based on GISAID column headers and use North America as a focal region--but these defaults might be removed in the future to remain data source and user base agnostic.

This workflow can optionally accept a "priorities" matrix so that the random sampling is biased towards the user's genomes of interest. However, a future version of this workflow should just accept the user's genomes of interest instead and compute the priorities matrix for them.

This PR additionally includes adjustments to the filter_subsample_sequences WDL task to convert the include_where and exclude_where input types from String? to Array[String]?, which better reflects how they would be used, and keeps a cleaner way of describing input values with spaces and special characters, which are better tolerated now.