Better output when evaluating student groups from Firebase dumps

• Generate scores-detail.txt, optimised for human readability
• Better statistics in scores*.csv: min, average, median and max per question
• Better header with question identifiers in scores*.csv
• Render scores with 2 decimal digits, and proper rounding
• Display uncaught runtime eval errors in stderr, instead of in eval.log