Mitigating optimistic bias in entropic risk estimation and optimization with an application to insurance