rqtl/qtl2: Version 0.20

Major changes scan1coef() and fit1() now, by default, gives coefficient estimates for the QTL effects that sum to 0, with an additional coefficient being the intercept. This makes it more like DOQTL (and scan1blup()). The previous behavior can be obtained with the argument zerosum=FALSE. Add function create_snpinfo() for creating a SNP information table from a cross2 object, for use with scan1snps(). Minor changes Updated extdata/mouse_genes_small.sqlite using updated MGI annotations. Some of the field names have changed. In check_cross2(), added a test for alleles being a vector of character strings. Fix some tests for R 3.6, due to change in random number generation. Use Markdown for function documentation, throughout Bug fixes In genoprob_to_snpprob() when a cross object is provided, make sure the genotype probabilities get subset to the cross markers. Fixed bug in scan1snps() re keep_all_snps=FALSE. It wasn't subsetting to the index SNPs properly. Added an internal function reduce_to_index_snps(). (See Issue #89.) Fixed bug in step probabilities for 4-, 8-, and 16-way RIL by selfing. Fixed bug in zip_datafiles() when the files are in a subdirectory. (See Issue #102.) Fixed bug in plot_peaks() for the case that the input peaks object does not contain QTL intervals. (See Issue #107.) Fixed inappropriate warning message for check of cross_info with cross type risib8. Fixed bugs in guess_phase() and locate_xo() where we needed an any() around a comparison of two vectors