Implemented option to Retrieve by paragraph

[andrelmfarias]

This option showed better performance on a smaller version of SQuAD-open (10% of questions). Here below are the results in comparison by Retrieve by document.

• Retrieve by document (top 3 documents): 53.1% (EM) / 56.8% (F1-score)
• Retrieve by paragraph (top 30 paragraphs): 61.4% (EM) / 66.1% (F1-score)

[codecov]

Codecov Report

Merging #252 into master will increase coverage by 2.27%.
The diff coverage is 62.5%.

@@            Coverage Diff             @@
##           master     #252      +/-   ##
==========================================
+ Coverage   34.06%   36.34%   +2.27%     
==========================================
  Files           8        8              
  Lines        1691     1695       +4     
==========================================
+ Hits          576      616      +40     
+ Misses       1115     1079      -36

Impacted Files	Coverage Δ
cdqa/reader/bertqa_sklearn.py	48.17% <ø> (+5.4%)	⬆️
cdqa/retriever/tfidf_sklearn.py	71.42% <ø> (ø)	⬆️
cdqa/pipeline/cdqa_sklearn.py	55.71% <54.54%> (+1.3%)	⬆️
cdqa/utils/converters.py	79.12% <80%> (-0.66%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 194d75f...aa45c29. Read the comment docs.