Consider requiring names to be unique

[LTLA]

Historically, names along the dimensions of an object are not required to be unique or non-empty, as those constraints can be violated when combining objects or subsetting them with duplicated indices.

By comparison, other names are required to be unique, e.g., assay names in an SE, column names in a DF. This is a tighter constraint that lends itself to more representations in other languages, e.g., Python dicts, JS objects.

Perhaps we can consider whether to force dimension-related names to be unique and non-empty. This would be mildly irritating to R users, who now have to sanitize their objects, but would provide more guarantees for readers.