Standardization of additional token identifiers across pipelines

`FluxPipeline` has utilities that give us `img_ids` and `txt_ids`:

https://github.com/huggingface/diffusers/blob/ce1063acfa0cbc2168a7e9dddd4282ab8013b810/src/diffusers/pipelines/flux/pipeline_flux.py#L514

https://github.com/huggingface/diffusers/blob/ce1063acfa0cbc2168a7e9dddd4282ab8013b810/src/diffusers/pipelines/flux/pipeline_flux.py#L385

As such these are not created inside the `transformer` class.

Whereas in `HiDream`, we have something different.

`text_ids` are created inside the `transformer` class:
https://github.com/huggingface/diffusers/blob/ce1063acfa0cbc2168a7e9dddd4282ab8013b810/src/diffusers/models/transformers/transformer_hidream_image.py#L796

`img_ids` are overwritten:
https://github.com/huggingface/diffusers/blob/ce1063acfa0cbc2168a7e9dddd4282ab8013b810/src/diffusers/models/transformers/transformer_hidream_image.py#L771C13-L771C20 (probably intentional because it's conditioned)

Then the entire computation

https://github.com/huggingface/diffusers/blob/ce1063acfa0cbc2168a7e9dddd4282ab8013b810/src/diffusers/pipelines/hidream_image/pipeline_hidream_image.py#L726-L744

happens inside the pipeline `__call__()`. Maybe this could take place inside a method similar to the `FluxPipeline`?

In general, these could be standardized a bit.

Cc: @yiyixuxu @a-r-r-o-w