Allow parameters of the ViTPooler to be configurable, with the default values set to the current hardcoded values

### Feature request I would like to add `pooled_output_size: int | None` and `pooler_act: str` fields in ViTConfig such that the following https://github.com/huggingface/transformers/blob/main/src/transformers/models/vit/modeling_vit.py#L665-L666 can be configured if needed. The two fields would have default values `None` (which results in using the value of `hidden_size`, like is currently done, and `tanh`. This would result in no change to the default behaviour, while allowing some flexibility. ### Motivation This is part of the open-sourcing of the HeAR model (https://arxiv.org/abs/2403.02522) in PyTorch format. The encoder uses the ViT architecture, and the pooler has no activation (identify function instead of tanh) and the pooling dimension is smaller than the hidden dimension. ### Your contribution I will submit a PR