Can Small LLMs Detect Defect-Prone Code Smells? An Empirical Evaluation of 8B-30B Models Against a Ground-Truth Dataset accepted at the AI Models/Data track of EASE’26.