Computes every student repository's per-question scores from the project's
grading database and writes them to scores.csv in the project root, one
row per repository with one column per question plus a total column.
Scores are recomputed the same way mark() displays them: the points of
the selected rubric items are summed (each item group's combined points
first clamped to its floor and ceiling) and passed through the question's
grading mode and score bounds. A repository/question pair that is not yet
graded (no selected rubric item and no public comment) is written as NA,
and a repository's total stays NA until all of its questions are
graded.