Removes the L0Learn dependency since it was no longer available on CRAN.

This version has some interface changes and minor improvements:

• susie_bhat and susie_ss are now merged to susie_suff_stat
• Additional input check and preprocessing options are implemented to susie_rss
• Add a method simple to estimate_prior_method that simply compares specified prior variance with zero and set it to zero if the latter has higher loglikelihood.

This version improves the sufficient statistics interface susie_ss and susie_bhat, and introduces a new model susie_rss for summary statistics version of SuSiE using RSS model. Specifically:

1. var_y is now replaced by yty in susie_ss
2. susie_z is now replaced by susie_rss using the new RSS model
3. Various checks for input covariance / correlation matrix of X, and for input summary statistics, to ensure they are in the space spanned by non-zero eigenvectors of covariance / correlation matrix of X
4. Add interface to access posterior quantities #84
5. Unify X data object for dense and sparse matrices

(Credits goes to @zouyuxin for 1-4 and @KaiqianZhang for 5)

This release is a patch to 0.7.0 that performs filtering by default with susie_get_CS function for effects with estimated V=0.