Track individual tensors in containers (datasets, tuples, etc)

[khatchad]

Currently, we track that datasets have tensors, but we don't track the tensors used to make the datasets. For example, the from_tensors() static method creates a dataset containing a single element corresponding to the passed argument. However, we don't encode this information in the dataset "created" in the analysis.

[khatchad]

In other words, it's a dataset of which tensors?

[khatchad]

This may be blocked on #42, as we won't be able to test this unless we can distinguish between the tensors using their shapes.