Support transparent inspection of logical problems when nested embedding composites used

Induced by #39, we want to support inspection of problems generated by multiple/nested embedding composites, in the implicit data collection mode.

Currently, only the outermost embedding is returned (see https://github.com/dwavesystems/dwave-system/issues/251), so we're unable to provide the correct (composed) embedding to the problem inspector.

A workaround for now is to explicitly provide the composed embedding (via [`compose_embeddings()`](https://github.com/dwavesystems/dwave-system/issues/251#issuecomment-579397013)).

The planned solution is to implement the dashboard-style inspection, where intermediate problems will be captured and inspected on demand.

Start by reviewing issue #39 and the linked discussion in issue #251, including the compose_embeddings() workaround. Then examine how implicit data collection currently exposes only the outermost embedding. Done means nested embedding composites can be inspected on demand with the composed embedding and intermediate problems captured.

Written by the indexing model from the issue text.