Consider adding a kwarg to skip checking whether a discrete constraint is valid

[arcondello]

When one calls CQM.add_discrete() we check that the given constraint is a valid discrete constraint - that is that it doesn't overlap with any other discrete constraint. The check happens in one of two places:

dimod/dimod/constrained/constrained.py

Lines 607 to 613 in e7bc4d5

	for v in variables:
	if v in self.variables:
	# it already exists, let's make sure it's not already used
	if any(v in self.constraints[label].lhs.variables for label in self.discrete):
	raise ValueError(f"variable {v!r} is already used in a discrete variable")
	if self.vartype(v) is not Vartype.BINARY:
	raise ValueError(f"variable {v!r} has already been added but is not BINARY")

or

dimod/dimod/constrained/constrained.py

Lines 671 to 683 in e7bc4d5

	for v, bias in qm.iter_linear():
	if v in self.variables:
	# it already exists, let's make sure it's not already used
	if any(v in self.constraints[label].lhs.variables for label in self.discrete):
	raise ValueError(f"variable {v!r} is already used in a discrete variable")
	if self.vartype(v) is not Vartype.BINARY:
	raise ValueError(f"variable {v!r} has already been added but is not BINARY")
	elif not vartype(v) is Vartype.BINARY:
	raise ValueError("all variables in a discrete constraint must be binary, "
	f"{v!r} is {vartype(v).name!r}")
	# we could maybe do a scaling, but let's just keep it simple for now
	if bias != 1:
	raise ValueError("all linear biases in a discrete constraint must be 1")

This is quite a slow check, and many users therefore bypass it by adding the discrete constraint as a constraint and then marking it as discrete.

label = cqm.add_constraint(x + y + z == 1)
cqm.constraints[label].lhs.mark_discrete()

This bypasses the check. Luckily, dwave-preprocessing handles fixing overlapping discrete constraints, so I think it's fairly safe to make the bypass more "legal", albeit with caveats explained clearly in the docs.

Therefore, we should/could consider adding a keyword argument skip_check defaulting to False that would allow users to bypass the check. If they then create overlapping discrete constraints they will be removed by presolve later.

cqm.add_discrete(["a", "b", "c"], skip_check=True)

For experience users who promise to construct the model without overlapping discrete constraints by default this is likely to be useful.