🚨 [v5] Rename left traces of past_key_value in BERT-like models

[zucchini-nlp]

What does this PR do?

As per title, finalizes the work from #41425 and renames past_key_value as past_key_values in all files without deprecation cycle

It was a find and replace, so tests might fail 😅

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: autoformer, bart, bert, bert_generation, bigbird_pegasus, biogpt, blenderbot, blenderbot_small, blip, bridgetower, camembert, data2vec, decision_transformer, electra, ernie, flaubert

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[vasqu]

Thanks a lot, all these big PRs made a mess :o lgtm

Just to be sure, can you check with run slow for the important models (bert, bart, gpt2, t5, roberta)

[zucchini-nlp]

Ran slow tests on Bert, BART, Ernie, GPT2, T5 and MusicGen comparing to main branch, all green ✅

[Cyrilvallez]

Not sure how they got back here because I had definitely made the switch in #39956 already... We need to be very careful when touching a lot of files otherwise we shoot ourselves in the foot!

[zucchini-nlp]

Yep, might be simply issues with rebasing main. Happens a lot when everyone is making huge PRs 😄

[vasqu]

Yup, sorry that's likely on me with the vLLM refactor 😢 wasn't aware of that rename