From recency to the central tendency bias in working memory: a unifying 1 attractor network model