Add ability to return chain break fraction per-qubit rather than per-sample

Right now, setting [chain_break_fraction](https://github.com/dwavesystems/dwave-system/blob/ecbb0267a4d6a7a0f7b54620471a8fc09b016841/dwave/embedding/transforms.py#L545) to `True` returns the fraction of chains broken in each sample.

Users are sometimes interested in knowing, across all samples, what fraction of times each qubit is in a broken chain. This would help identify if an individual penalty strength needs to be higher.