Add TF auto model to the docs + fix sphinx warnings (again)

[sgugger]

The goal of this PR was just to add the automodel to the docs (as pointed out in #5145) but then I had 100 sphinx warnings that made me grumpy so I had to fix them (some of them from the auto model docs, but most of them from #4978).

As usual a few of them were harmless, others had a real impact on the docs, so I think we should make sure in the CI that there are no sphinx warnings to avoid making the docs bad by mistake.

[codecov]

Codecov Report

Merging #5187 into master will decrease coverage by 0.89%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #5187      +/-   ##
==========================================
- Coverage   77.99%   77.10%   -0.90%     
==========================================
  Files         138      138              
  Lines       23772    23772              
==========================================
- Hits        18541    18329     -212     
- Misses       5231     5443     +212     

Impacted Files	Coverage Δ
src/transformers/modeling_albert.py	80.43% <ø> (ø)
src/transformers/modeling_bart.py	96.23% <ø> (ø)
src/transformers/modeling_bert.py	87.50% <ø> (ø)
src/transformers/modeling_ctrl.py	99.14% <ø> (ø)
src/transformers/modeling_distilbert.py	97.70% <ø> (ø)
src/transformers/modeling_electra.py	80.19% <ø> (ø)
src/transformers/modeling_flaubert.py	84.12% <ø> (ø)
src/transformers/modeling_gpt2.py	85.43% <ø> (ø)
src/transformers/modeling_longformer.py	93.04% <ø> (ø)
src/transformers/modeling_mobilebert.py	88.78% <ø> (ø)
... and 24 more

---

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 ebc3610...cca9865. Read the comment docs.

[LysandreJik]

Very cool! Thanks for taking care of the warnings, I'm opening a PR for the test failing on error right now.