Fix convert_token_type_ids_from_sequences for fast tokenizers

[n1t0]

Before this fix, the generic version of convert_token_type_ids_from_sequences from tokenizer_utils gets called when called on a PreTrainedTokenizerFast. The type_ids for the special token are thus not included.
There is no way at the moment to get this information from the rust tokenizers, so we just use the implementation from the original python tokenizers. Tests added as well.

Thanks @dirkgr for reporting this.

[codecov-commenter]

Codecov Report

Merging #4503 into master will increase coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #4503      +/-   ##
==========================================
+ Coverage   77.83%   77.86%   +0.02%     
==========================================
  Files         123      123              
  Lines       20514    20526      +12     
==========================================
+ Hits        15968    15982      +14     
+ Misses       4546     4544       -2     

Impacted Files	Coverage Δ
src/transformers/tokenization_bert.py	95.00% <100.00%> (+0.12%)	⬆️
src/transformers/tokenization_roberta.py	94.52% <100.00%> (+0.49%)	⬆️
src/transformers/modeling_tf_utils.py	88.66% <0.00%> (+0.16%)	⬆️
src/transformers/file_utils.py	73.85% <0.00%> (+0.41%)	⬆️

---

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 a086527...795f44a. Read the comment docs.

[LysandreJik]

LGTM, thanks @n1t0