Fix duplicate outcomes

[kuchenrolle]

Changes proposed in this pull request:

• outcomes did not get filtered for duplicates with remove_duplicates=True, but they do now. consequently the comparison to the reference file fails, because that one contains duplicate outcomes.
• re.split() removes the split pattern, so process_context() did not do anything. i removed it and simplified (i think) that part of the code.
• gen_occurrences() generated an empty occurrence, i changed that by turning the while loop with ifs into a for loop (which is more similar to how "word_to_word" is handled), but alternatively the break condition could just be moved up two lines.
• handling of "bigrams_to_word" and "trigrams_to_word" only differs in one parameter, so i removed the repetition.
• i don't understand why the diff shows the temporary_directory, that should be the same in both branches. sorry about that.

[coveralls]

Coverage decreased (-0.06%) to 83.274% when pulling d274a8f on fix-duplicate_outcomes into 9b47606 on master.

[derNarr]

• add test that compares remove_duplicates=True to remove_duplicates=False in preprocess.py