If a word never occurs in training, it may or may not participate in vector norm calculations.

Currently, it depends on whether the word is in 'DataFactory::wordRestrict' list.

'wordRestrict' list contains ONLY TRAINING words when:
- separate Lemur index for training and testing is used
- plain file is used for input data

'wordRestrict' list contains BOTH TRAINING AND TEST words when:
- single joint Lemur index for training and testing is used

//TODO: in case of joint Lemur index, collect de facto words for 'wordRestrict'
