Can an automated three-dimensional statistical shape model increase component rotation measurement validity to better explain total knee arthroplasty patient dissatisfaction?