Problems with CosineSimilarity implementation

[Torsten Zesch]

Dear users,

recently, some serious issues with the implementation of
CosineSimilarity have surfaced.

I tried to fix that in the recent snapshot, but it would be good if
you could report any problems that might occur with the new version.

I also removed some questionable choices, like pur IDF weighting which
makes little sense, as both document vectors need to be identical
then.
If you have been using this productively, please get in touch and
convince me when this is actually needed ;)

Also note that all experiments that have been using this measure
before probably give wrong results too.

sorry for the inconvenience,
Torsten