fix(utils): Resolve backbone utils test regressions

[harshaljanjani]

What does this PR do?

The following regressions were identified and fixed in this PR:

→ 0c4fe5c ("Delete duplicate code in backbone utils") removed use_pretrained_backbone from load_backbone; pretrained weights are now loaded via from_pretrained on the parent. The tests in test_backbone_utils.py#L217 weren't updated likely because tests/utils/ wasn't in the run-slow scope of that PR. Updated the test to match the new design (load_backbone always initializes with random weights).
→ dff54ad ("out_indices always a list") made out_indices always return a list via backbone_utils.py#L57. The assertion at test_backbone_utils.py#L165 was written before that change in #28784 and still expected a tuple. Updated the assertion to expect a list.

cc: @zucchini-nlp @Rocketknight1

CI coverage fixed by this PR (checked in the latest CI run):

tests/generation/test_utils.py::GenerationIntegrationTests::test_green_red_watermark_generation, tests/utils/test_backbone_utils.py::BackboneUtilsTester::test_load_backbone_from_config, tests/utils/test_backbone_utils.py::BackboneUtilsTester::test_load_backbone_in_new_model

Current output:

Output after fix:

Code Agent Policy

• I confirm that this is not a pure code agent PR.

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 fix any necessary existing tests?

[Rocketknight1]

cc @ydshieh @tarekziade maybe? We may want to consider ditching these tests if they're all slow and we don't even notice when they break!

[zucchini-nlp]

Hm, I usually try to check daily CI for things that I might have broken, so prob this one slipped off the CI-bot feedback? Been quite a while since the last time "timm" was updated...

[harshaljanjani]

Good day; just checking in to see if there are any updates on the direction we should take regarding this :)

[zucchini-nlp]

sorry for delay, I checked with the prev breaking PRs and existing tests. I think these slow tests are meant to be in timm_backbone tester as IntegrationTests, whenever it makes sense

https://github.com/huggingface/transformers/blob/main/tests/models/timm_backbone/test_modeling_timm_backbone.py

Since we simplified a lot loading backbones, and there's no specific loading anymore, some could be not-useful anymore. For ex test_load_backbone from config/kwargs/ckpt are tested in individual models

[harshaljanjani]

Will look into the same and revert, thank you :)