Fix token/label offset issue

[ivyleavedtoadflax]

• Solves issue with prodigy_to_tsv which was causing differences between output labels and tokens.
• Adds new tests and test data (this is why the PR is so big)

[codecov-io]

Codecov Report

Merging #22 into master will decrease coverage by 1.94%. The diff coverage is 70.65%.

@@            Coverage Diff             @@
##           master      #22      +/-   ##
==========================================
- Coverage   84.56%   82.61%   -1.95%     
==========================================
  Files          16       16              
  Lines        1192     1214      +22     
==========================================
- Hits         1008     1003       -5     
- Misses        184      211      +27     

Impacted Files	Coverage Δ
deep_reference_parser/train.py	0.00% <0.00%> (ø)
deep_reference_parser/prodigy/prodigy_to_tsv.py	88.23% <83.07%> (-9.25%)	:arrow_down:
deep_reference_parser/io/io.py	97.14% <100.00%> (+0.26%)	:arrow_up:

---

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 bb49ede...27648b1. Read the comment docs.