Consider rethinking IDs

[tkw1536]

Suggestion by @florian-rabe:

Every dataset should have

• an id (ideally a user-chosen short string)
• a set of properties, whose concatenated string renderings define unique identifiers for the items.

The former should be the primary id of the dataset. The pair of those two should be the id of each item. Every property should have an id that is unique within the dataset.

The triple of of dataset, item, and property id should be the id of each datum.

These id's should be used both for internal references across datasets and for citations from the outside.

[tkw1536]

The current behavior is that each item gets an RFC4122 random UUID. It is guaranteed unique across MathDataHub. These UUIDs are a direct consequence of our table structure and have no meaning attached.

I agree that we might want author-defined local IDs of items within datasets. I am not sure if those should be the primary ID used by the system.