remove the nltk POS tagger from convert_rst_discourse_tb.py

[mheilman]

Currently, convert_rst_discourse_tb.py uses NLTK's POS tagger to create flat trees for sentences that are in the RST treebank but not the Penn Treebank. This dependency should eventually be removed and replaced with ZPar.

[YTZ01]

Hello, I have a problem in running this line'''convert_rst_discourse_tb ~/corpora/rst_discourse_treebank ~/corpora/treebank_3'''. I'm wondering the PDTB dataset in your setting is PDTB-v1（2019） or PDTB-v2（2020）, cause I downloaded the dataset from LDC, but it doesn't have a 'parsed' file under it, only data，docs and tools, index.html. Have you met this issue? @mheilman