[Docs] Add Developer Guide: How to Hack Any Transformers Model

[MagnusS0]

What does this PR do?

This PR adds a new developer guide titled "How to Hack Any Transformers Model" to the docs. The guide shows how to modify existing models, using the Segment Anything Model (SAM) as an example. It also encourages community contributions by inviting others to share their own hacks.

Changes

• Added a new developer guide at docs/source/en/how_to_hack_models.md
• Updated docs/source/en/_toctree.yml to include the new guide in the Developer Guides section.

Fixes #33928

Before submitting

• This PR improves the docs.
• I have read the contributor guideline.
• This was discussed in the GitHub issue Separate q_proj, k_proj, and v_proj for Attention Layers in SAM #33928.

Who can review?

Tagging @ArthurZucker as we discussed this in issue #33928 and the code example was provided by him. Let me know if this is what you had in mind 🤗

[ArthurZucker]

Nice! I think we can have a separate doc, with sam being the example!
This could be in the same section as https://huggingface.co/docs/transformers/fast_tokenizers (developer guides!)

[MagnusS0]

Ahh, yeah makes sense, then I can extend the example and show how it also can be used e.g. with PEFT! WDYT?

[ArthurZucker]

Of course! And if people have example of adding SDPA for example (not here for SAM) or good hacking, will go there! let's call for contribution probably! 🤗

[MagnusS0]

Updated the PR: It now adds a new developer guide titled "How to Hack Any Transformers Model" 🚀
I've also updated the PR description and title accordingly.

Let me know your thoughts! Do you think we should make the guide more general regarding model hacking? Or do you (or anyone else) have any extra examples to add?