sample_ising may drop variables when h is a list

[mcfarljm]

For example, using a structured sampler where qubit 0 is not available, the following silently drops the variable and produces an empty sampleset:

DWaveSampler().sample_ising([1], {})

whereas using a dict for h that includes an invalid qubit (or as part of J) will raise an error:

DWaveSampler().sample_ising({0: 1}, {})

BinaryQuadraticModelStructureError: Problem graph incompatible with solver.

Documentation for the base class method is:
https://github.com/dwavesystems/dimod/blob/25065ad8d30154098f55d827022b815421822864/dimod/core/sampler.py#L200-L205

Possibilities for addressing this include:

• Deprecating the use of a list for h in sample_ising (altogether or just for DWaveSampler?) Con: may still be value in allowing it from memory efficiency perspective.
• Propagate a warning
• Raise an exception. This would be more consistent with the behavior when h is a dict, and it matches the docstring of the original behavior.

Original docstring:

        *h* --- List or tuple of the linear Ising coefficients. The
         :math:`h` value of a non-working qubit must be zero or an
         exception will be raised. Inactive qubits are disabled during
         annealing and cannot distinguish their states.

Relevant code in DWaveSampler:

dwave-system/dwave/system/samplers/dwave_sampler.py

Lines 365 to 375 in bedfe51

	def sample_ising(self, h, *args, **kwargs):
	# to be consistent with the cloud-client, we ignore the 0 biases
	# on missing nodes for lists
	if isinstance(h, list):
	if len(h) > self.solver.num_qubits:
	msg = "Problem graph incompatible with solver."
	raise BinaryQuadraticModelStructureError(msg)
	nodes = self.solver.nodes
	h = dict((v, b) for v, b in enumerate(h) if b and v in nodes)
	return super().sample_ising(h, *args, **kwargs)