Fix 289

[bertsky]

@fstrunz @andbue sorry I'm late with this, but #289 is actually still broken: the skip setting does not work.

It will always raise if some pred files are missing.

The fix actually requires two things:

• if skip is set, the superfluous GT ids must be ignored
• at the end of the evaluation, at the time the confusion table and worst examples are to be reported, the existing code constructed the dataset ad-hoc anew (but without the skip rules), ending up in a similar exception. To avoid this, the set of GT IDs must be kept and passed along to the very end

[andbue]

Thanks for sorting this out!