There's now results that indicate errors in addition to the regular good
results. This updates generate so it actually parses the lines into
Result* structures and therefore can easily ignore the ones that have a
ResultError* type.

Instead of scaling the lines we have such that they add up to the given
total (really bad if we've only done a partial scan), give ourselves a
budget of size CONSTANT * NUM_LINES. Now partial scans are more
meaningful and all scanners' numbers should be directly comparable.

    note things to document further or change