better warnings based on r_ai_ss

Plot correlation between questions.

check that answers are distinct

also check that question variants are distinct?

More stats to report: m_{QVA}(q,v,a) = average score of all students

answering $a$ for variant $v$ of question $q$. ? Could also report

m_{QVB}(q,v,b) = average score of all students getting question $q$,

variant $v$ correct (if b=1) or incorrect (if b=0).

Integrate bad-key error reporting (closest Hamming match, best score match).

Check for repeated exam codes.

Add P_sq(s,q), c(s,q), a(e(s),q) to scores.csv so that we can see what

each student answered and what the correct answer should have been.

Fix bug reported by Lee on 2013-02-14 where blank answers are

converted to * which are then interpreted as answers by n, na,

statistics, etc, leading to these statistics thinking every student

answered every question.