Clarify extension usage validation

Currently, the use of extensions is validated in (what I first thought to be) a straightforward way. One aspect of that validation is:

- When there is an extension found during the traversal (for example, a `boundingVolume.extensions["3DTILES_bounding_volume_S2"]`), then the `"3DTILES_bounding_volume_S2"` is stored as a "found" extension in the validation context
- At the end of the validation, the "found" extensions are compared to the `extensionsUsed` that have been declared. When an extension was "found" but not declared as "used", a warning is created. When an extension was declared as "used", but not "found", a warning is created.

The problem (which is obvious in hindsight): Certain extensions are never "found" explicitly. For example, `3DTILES_content_gltf`. It never appears as an object anywhere, but only _allows_ glTF/GLB files to be used as tile content.

This can be solved, to some extent, by keeping track of the _actual_ content data types that have been encountered. (There are corner cases where this becomes less trivial. For example, when the content data validation is disabled via a command line flag).

A related question is the _version dependency_. A specific example is that `3DTILES_content_gltf` should probably not be used in a `1.1` tileset, and using it should cause a warning. The more general question is: How to encode the mix-and-match between "tileset versions" and "allowed extensions". There may be extensions that are valid only in `1.0`, other may be valid only in `1.1`, and others may be valid in both...

Trace how traversal records found extensions and how extensionsUsed warnings are produced, then inspect how content data types and the content-validation CLI flag affect that context. Define the rules for implicit extensions such as 3DTILES_content_gltf, including tileset-version compatibility, and finish with tests covering undeclared, unused, disabled-validation, and version-dependent cases.

Written by the indexing model from the issue text.