Fixing missing arguments for TransfoXL tokenizer when using TextGenerationPipeline

[TevenLeScao]

As discussed with @LysandreJik and @mfuntowicz , TextGenerationPipeline gives imperfect results when using TransfoXL as the tokenizer lacks the add_space_before_punct_symbol argument. In order to fix this, this PR overrides _parse_and_tokenize for this pipeline in order to pass tokenizer arguments.

[codecov]

Codecov Report

Merging #5465 into master will decrease coverage by 0.12%.
The diff coverage is 85.71%.

@@            Coverage Diff             @@
##           master    #5465      +/-   ##
==========================================
- Coverage   77.52%   77.40%   -0.13%     
==========================================
  Files         141      141              
  Lines       24610    24617       +7     
==========================================
- Hits        19079    19054      -25     
- Misses       5531     5563      +32     

Impacted Files	Coverage Δ
src/transformers/pipelines.py	76.00% <85.71%> (+0.11%)	⬆️
src/transformers/tokenization_albert.py	28.84% <0.00%> (-58.66%)	⬇️
src/transformers/generation_utils.py	96.82% <0.00%> (-0.29%)	⬇️
src/transformers/file_utils.py	79.59% <0.00%> (+0.33%)	⬆️
src/transformers/configuration_utils.py	96.40% <0.00%> (+0.71%)	⬆️
src/transformers/tokenization_auto.py	97.77% <0.00%> (+2.22%)	⬆️
src/transformers/tokenization_xlm_roberta.py	95.23% <0.00%> (+33.33%)	⬆️

---

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 6726416...25f8c86. Read the comment docs.

[mfuntowicz]

LGTM! 👍