Fix per-layer rope_parameters misparsing when layer_types omits keys

[Noxtimo]

What does this PR do?

Fixes #48392.

Problem

Models that use heterogeneous attention layers (e.g. Gemma3, Olmo3 mixing sliding and full attention) populate rope_parameters with entries for every possible layer type (sliding_attention and full_attention). However, when a model configuration is instantiated with fewer layers than the sliding window period (such as Gemma3TextConfig(num_hidden_layers=2) or Olmo3Config(num_hidden_layers=2)), config.layer_types contains only ["sliding_attention", "sliding_attention"].

Previously, standardize_rope_params and validate_rope checked set(rope_parameters.keys()).issubset(layer_types). When not all layer types were instantiated, this check returned False, causing the nested dictionary to be misclassified as a single flat config dict:

1. Emitting a spurious warning: Unrecognized keys in rope_parameters for 'rope_type'='default': {'full_attention', 'sliding_attention'}.
2. Polluting config.rope_parameters with top-level "rope_type": "default" and "rope_theta": None which was then persisted to disk on save_pretrained.
3. Bypassing validation of actual per-layer RoPE parameters (e.g., yarn, linear) for non-instantiated layer types.

Solution

• In src/transformers/modeling_rope_utils.py, validate that rope_parameters keys are a subset of allowed_layer_types (checking _rope_type_labels if present, or ALLOWED_LAYER_TYPES from configuration_utils.py) rather than only currently instantiated self.layer_types.
• In standardize_rope_params, iterate over set(rope_parameters.keys()) so all configured layer types are standardized.
• Added comprehensive unit tests in tests/utils/test_modeling_rope_utils.py verifying clean standardization, warning suppression, validation, and save_pretrained/from_pretrained round-tripping for partial layer configurations.

---

Before submitting checklist

• 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: Fixes Nested per-layer rope_parameters is misparsed when layer_types omits one of its keys #48392
• 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?

@zucchini-nlp @ArthurZucker @Cyrilvallez

[github-actions]

---

CI recap

Dashboard: View test results in Grafana
Latest run: 33267969929:2
Result: success | Jobs: 16 | Tests: 179,213 | Failures: 0 | Duration: 14h 32m