Improve MbioDataset validation

[d-callan]

now that we can accept UDs as treeSE and make MbioDatasets from them.. were seeing cases where some things unclassified at say the genus level appear as empty strings in the genus collection resulting from using importTreeSummarizedExperiment. we should both handle this case better so that (similar to Incertae Sedis in our curated datasets) we produce the known lineage for these taxa rather than empty strings so we can at least give people what we do know.. also, the validation for the Collection class should make sure all of its column names are unique.