Support multiple `timm` as backbones with `forward_intermediates` method

### Feature request Currently, in `TimmBackbone` requires that the `timm` model has a `return_layers`, which I'm not sure if any model implements: https://github.com/huggingface/transformers/blob/c15aad0939e691d2ffdbac7ae71921b51fe04e3f/src/transformers/models/timm_backbone/modeling_timm_backbone.py#L80-L83 Lately, `timm` implemented a `forward_intermediates` method to almost all networks for feature extraction - [issue](https://github.com/huggingface/pytorch-image-models/issues/2131) and [pr](https://github.com/huggingface/pytorch-image-models/pull/2136). ### Motivation Thus, will little changes, we will be able to do ```py import transformers model = transformers.Mask2FormerModel( config=transformers.Mask2FormerConfig( use_timm_backbone=True, backbone="vit_small_patch16_224", ), ) ``` as currently we get the error: ```sh AttributeError: 'FeatureGetterNet' object has no attribute 'return_layers' ``` ### Your contribution Review and/or implemented the PR