Discrepancy between EmbeddingComposite's chain strength and uniform_torque_compensation's chain strength for QUBO problems

**Current Problem**
When using the EmbeddingComposite, the BQM is always converted to Ising and the chain strength is calculated with Ising biases. This chain strength is then saved in `sampleset.info['embedding_context']['chain_strength']`.

If the user calls `uniform_torque_compensation` (or any of the chain strength functions), with a QUBO problem, the chain strength they get will be different from the one that is saved (because it now uses QUBO biases). This could be confusing for the user.

**Proposed Solution**
One possible solution would be to save the `smear_vartype` onto `embedding_context` as well, so that the user can see whether the chain strength was calculated with Ising or QUBO biases in mind.