How to Reconstruct Accurate Phylogenetic Trees from Nucleotide Sequence Data with Extraordinary Compositional Bias: Assessment of the Performance of Data-Recoding Methods and Non-Homogeneous Models