fix: confidence of a one-option distribution is 1.0, not a ZeroDivisionError - #11

confidence(p) is defined as 1 - H(p)/ln(K). When p has one entry, ln(1) is 0 and the call raises:

>>> from openjev.engine import confidence
>>> confidence([1.0])
ZeroDivisionError: float division by zero

An empty list hits the same division. One-option Choice/Score answers are currently short-circuited as "forced" answers before confidence runs, so the crash is latent, but the helper is part of the public engine API (imported and tested) and a single-outcome distribution is fully certain by definition.

Change: return 1.0 for a one-element distribution and 0.0 for an empty one, matching the docstring ("1 is certain, 0 uniform"). Existing multi-option cases are unchanged.

Test: test_confidence now asserts confidence([1.0]) == 1.0 and confidence([]) == 0.0. Both raise ZeroDivisionError on main and pass with this change.

python3 -m pytest -q tests/test_api.py::test_confidence   # 1 passed