`snp_unphased` and `snp_phased_ancestry` revamp

[JamesYang007]

• nnm(j) to return number of non-missing values for each column.
• imputed_value(j) to return imputed value for each column.
• write should take in impute method.
• modify docs
• benchmark compute_column_mean, compute_nnm, compute_nnz.
• document that negative entries will be saved as some implementation-defined negative value, so to_dense may not recover the exact negative value from before.
• redo representation slightly for better cache
• smart cov calculation with optimization for when i == j case.
• change order of operation for muls to be category then columns.
• check calldata only contains 0/1 (no missing values!) and ancestries has values [0, A).