Should permit nesting models without links (e.g. defining frames / visuals / collision only)?

[EricCousineau-TRI]

In SDFormat 1.7, we admit models with no links iff they are //static=true, and the canonical link for those models are attached to the world:
http://sdformat.org/tutorials?tut=pose_frame_semantics_proposal&cat=pose_semantics_docs&#2-1-implicit-frame-defined-by-model-pose-attached-to-canonical-link

However, @azeey pointed out that we may want similar behavior for nested models:
gazebosim/sdf_tutorials#34 (review)

[EricCousineau-TRI]

Thoughts:

• Not sure how I personally feel about just adding frames / visuals / collision without inertias. (if you're adding new collisions via SDFormat, they should be real and physical?)
• Not sure how I feel about this when the adding element will have scoped names.

So thinking about this, my gut says we punt for now on SDFormat 1.8.

[EricCousineau-TRI]

This should've been accomplished for static models per latest PRs from @azeey

@brawner to confirm that this is already covered in a unittest.

[brawner]

Resolved via #341 and is tested in the model_dom.cc integration test DOMRoot.LoadNestedCanonicalLink and DOMRoot.LoadNestedExplicitCanonicalLink (https:/osrf/sdformat/blob/5a910816ef57ebd34aa4cf0c0a233750d8af88b3/test/integration/model_dom.cc#L433-L531)