Refine the split of a dominant module #55

Since #54, a module that holds more than 70 % of the universe files (and at least 20 files) is split into its directories. Two follow-ups:

• Measure dominance over non-test files. Test files count toward the share today. In a common layout, src/ holds 55 % of the files, a mirrored test/ 40 % and scripts/ 5 %. No module passes 70 %, so src stays whole and its cohesion says nothing, while test is test-only and not ranked. Counting only non-test files would split src here. That changes which repositories are split, so the threshold needs the same evidence as Split a module that holds most of the code, even when small modules exist #31 (this repository, angular/angular, several single-package repositories).
• Leakage and depth for a split package. After a split, the entry file usually forms a module of its own, so leakage and depth of the split code are null. They could still be measured on the declared package (its manifest entry points against all of its files), separately from the directory modules used for cohesion.

Done when: each item is either implemented with tests and measured before and after, or explicitly declined in this issue.