Reject quadratic biases for unknown edges

[danielgaskins]

While looking at #19, I noticed that set_quadratic leaves the edge index as None when neither orientation exists in the model. PyTorch treats parameter[None] as an extra dimension, so assigning the requested bias broadcasts it across every quadratic weight instead of rejecting the input.

Raise a ValueError before the assignment. The existing test happened to use a nonexistent edge and only checked the first overwritten bias, so update it to use an actual reversed edge and add a regression test confirming that an unknown edge leaves every bias unchanged.

Tests:

• python3 -m unittest discover -s tests -v
• reno lint
• python3 -m compileall -q dwave tests