Fix VariableExactSumConstraint forwardcheck incorrectly pruning valid domain values - #108

Summary

VariableExactSumConstraint.__call__ (no-multipliers path) builds
sum_value by accumulating assigned variable values plus
min(domain) as a placeholder for each unassigned variable. In the
forward-check loop it then computed:

temp_sum = sum_value + value

This double-counts the placeholder for the variable whose domain is
being pruned, making temp_sum too large and hiding valid values.

Minimal reproducer (returns [] instead of [{a:1,b:2,c:3}, {a:2,b:1,c:3}]):

from constraint import Problem, VariableExactSumConstraint
problem = Problem()
problem.addVariable("a", [1, 2])
problem.addVariable("b", [1, 2])
problem.addVariable("c", [3])
problem.addConstraint(VariableExactSumConstraint("c", ["a", "b"]))
print(problem.getSolutions())   # [] — wrong

With a=1 assigned and c=3 assigned, sum_value is 1 + min([1,2]) = 2.
The check 2 + 2 = 4 > 3 then hides b=2, even though a + b = 1 + 2 = 3
satisfies the constraint.

Fix

Subtract the placeholder for the variable being checked before adding the
candidate value:

var_placeholder = min(domain)
for value in domain[:]:
    temp_sum = sum_value - var_placeholder + value
    if temp_sum > target_value:
        domain.hideValue(value)

The multipliers branch was not affected because it never adds a
placeholder (it only tracks missing = True without modifying sum_value).

Tests

All 34 existing constraint/solver tests continue to pass. The note
about test_if_compiled failing when .so files are absent is a
pre-existing CI condition unrelated to this fix.

This pull request was prepared with the assistance of AI, under my direction and review.