[Gemma] Modular-friendly buffers

[RyanMullins]

What does this PR do?

Per a conversion with @Cyrilvallez on Slack on how to make Gemma models more compatible with modular inheritance, this PR:

• Updates Gemma models to use nn.parameter.Buffer instead of a self.register_buffer() for static attributes
• Updates _init_weights functions as appropriate to populate non-persistent buffers correctly

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@ArthurZucker @Cyrilvallez

[github-actions]

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

run-slow: cwm, dots1, exaone4, exaone_moe, gemma, gemma2, gemma3, gemma3n, gpt_oss, granitemoehybrid, lfm2, lfm2_moe, minimax, ministral, olmo3, olmo_hybrid

[Cyrilvallez]

Arf sorry, maybe I wasn't clear enough offline 🥲 Unfortunately, nn.parameter.Buffer was only added in torch 2.5, and we still support 2.4... So we need to hold off at least a bit for this PR! Let me see internally if we can bump the dependency so we can use this!