Define ideal compartment and metadata table relationships for joins to avoid returning NA results

@jenna-tomkinson brought up a scenario in which cytoplasm and cells table data may exist in a dataset where nuclei may be missing. As-is, this could produce rows of CytoTable output which include cytoplasm and cells feature data but NA's for nuclei data. This happens because the cytoplasm table is used as a basis for the preset joins (there's no constraint that would prevent someone from specifying otherwise). We could use the nuclei table as the basis for preset joins which would remove this possibility. Chatting about this together with @MikeLippincott it came up that retaining the NA rows could be important information for a data scientist to see as part of their efforts.

Opening this issue up to help document the discussion and provide an area for further thoughts to be documented.