Identification of linear relationships from noisy data using errors-in-variables models - relevance for reconstruction of past climate from tree-ring (and other) proxy information