Revert "references"

[danielballan]

Inspired by conversations with metadata experts at NIST, I think we should declare #242 a failed experiment. It was added to meet a perceived urgent need, but in practice was not yet much used. Removal is expected to have a very low (possibly zero) impact on our early users.

As @dylanmcreynolds pointed out at the time in his comments on #242, there are existing standards for placing references inside a JSON payload. Promoting a Tiled-specific area outside metadata is, I think, unhelpful added complexity. It is better to use these standards within the metadata.

I notice that this gives a nice symmetry to the Tiled data node:

• structure and structure_family
• metadata and (optional) specs

Each of these pairs provide information then a description of how to interpret that information. These, along with a path that locates the note within the logical tree layout, fully describe Tiled's data model.

Promoting additional items like references, time_created, and time_updated into the API as first-class items alongside metadata feel like a slippery slope toward re-inventing a data model. Better to rely on existing standards, use them within metadata and (optionally)declare them withinspecs`.

Note: For forensic reasons, I think we should track time_created and time_updated in the database but probably not expose it in the HTTP API. If time matters, it should be in metadata and should refer to a semantically important time (like experimment time) related to the data, not database book-keeping.

[padraic-shafer]

This seems like an elegant way handle metadata & specs.