Reliability of a laboratory-based long sprint cycling test: applications of the smallest worthwhile changes in performance for repeated measures designs