Review which data fields we're actually using

The model structs we currently have closely follow the data sent by tiled but much of it is not used or not reliable.

* Search responses have a "meta" field with optional and arbitrary data
* RunMetadata has a "start: Start" field mapping to the start document but this is never used in glazed (and can't be relied on to be anything more than an arbitrary json value).
* Attributes has "access_blob", "specs",
* Links contains a combination of all the link fields used by any of the types in tiled and might be better served by a hashmap
* etc

It might be useful to remove the fields we're not using and replacing them if and when they're being used in glazed.

It would make testing a lot easier when we don't have to return a vast amount of unrelated data from mock servers to test query handling.

Start by inspecting the model structs mentioned in the issue—RunMetadata, Attributes, and Links—and trace their use in query handling and mock server tests. Done means identifying which fields are unused or unreliable, deciding what should remain or change, and updating affected tests and mock responses.

Written by the indexing model from the issue text.