Feature Request: report bootstrap probabilities across hierarchy even when flat mapping is used

[berl]

When flat mapping is used, the bootstrapping probability results at the finest resolution of the hierarchy accurately report the fraction of time a cell is mapped to the most-mapped category. This same probability is currently reported across all levels of the hierarchy. At other levels of the hierarchy, however, it's not accurately reflecting the fraction of bootstrap trials that matched the most-mapped category at that level. This would provide useful information about the stability of mapping results through the levels of the hierarchy. For example, in a reference with Subclass and Cluster levels, bootstrapping across clusters resulting in a Cluster_bootstrapping_probability == 0.5 could stay entirely within the parent Subclass (Subclass_bootstrapping_probability == 1.) or the bootstrapping results could span multiple Subclasses (say, Subclass_bootstrapping_probability == 0.6).

These hierarchical bootstrap results could help users understand how mapping quality and stability is distributed across cells and across a reference taxonomy.