[docs] modular transformers

[stevhliu]

refactors the how to add a model with modular transformers doc:

• structure:

  • flipped the order so you learn how to write the modular file first before generating it
  • remove the motivator examples with BERT/RoBERTa
  • merge the two super() sections with context on when to use each
  • replace full code blocks with diffs so its easier to see whats actually changing

• missing features:

  • @strict, __post_init__, __all__, AttributeError(), base_model_tp_plan, base_model_pp_plan

• updates:

  • update the### Config section with new patterns like __post_init__, AttributeError() and class-level type annotations for new attributes
  • super_kwargs example references GemmaForCausalLM now
  • adds the modular-detector-v2 Space

let me know if there are other commonly missed modular practices!

todo:

• add link to [docs] vlm addition #45271 once merged on how to add vision processing components
• add link to [docs] model testing #45152 once merged on how to write model tests and required PR checks
• add link to [docs] @auto_docstring decorator #45130 once merged on how to use @auto_docstring decorator

[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.

[zucchini-nlp]

nice, i was pointed out just recently that docs aren;t up-to-date with config changes 😄

[casinca]

Hi @stevhliu , if I may, would it be possible to add somewhere what Matt said in #45141 (comment).
I believe this would help others that might end up in the same situation.