This PR implements preliminary code for within group knockoff filter, a desired featured noted in #1.
Knockoff discoveries are still based on the group-level W
But we do store and output the variant-level W
When making manhattan plots, all SNPs are plotted by their group-W, but we label a SNP based on the variant-level W. This could lead to interesting behavior where a SNP with larger group-level W has weaker variant-level W, as shown in step 5 of full example tutorial in the documentation
This PR implements preliminary code for within group knockoff filter, a desired featured noted in #1.