Major changes:
- The main error model switched from a simple Kernel Density Estimation to a two-dimensional Kernel Density Estimation. The new model is better at estimating the mean sequence quality and produces more realistic sequences overall
- The insert size of the generated pairs is no longer fixed but follows a more realistic distribution
- added a `--gc_bias` option to naively discard some sequences that fall outside a normal range of GC content.
Bug fixes:
- fixed a bug that would affect the substitution rate for T, G and Cs