Diffusers-interpret

- Add token attributions visualization; kudos to TomPham97 👏
python
output.token_attributions.plot(normalize=True)

- `output.token_attributions` is now a list [`TokenAttributions`](https://github.com/JoaoLages/diffusers-interpret/blob/83518e5f1691c2b3f4fc1c656f97eed3a3413a6f/src/diffusers_interpret/token_attributions.py#L5) object and `output.pixel_attributions` is now a np.ndarray [`PixelAttributions`](https://github.com/JoaoLages/diffusers-interpret/blob/83518e5f1691c2b3f4fc1c656f97eed3a3413a6f/src/diffusers_interpret/pixel_attributions.py#L8) object (both compatible with previous version)
- `output.normalized_token_attributions` and `output.normalized_pixel_attributions` now raise a DeprecationWarning (but both still work). Use `output.token_attributions.normalized` and `output.pixel_attributions.normalized` instead.

- Full compatibility with `StableDiffusionImg2ImgPipeline` and `StableDiffusionInpaintPipeline` 🚀
- New GIF visualization for all the generated images during the diffusion process (kudos to TomPham97 )
- Cleaner API, docstrings and usage documentation

fix `n_last_diffusion_steps_to_consider_for_attributions=0`

Full compatibility with [diffusers v0.3.0](https://github.com/huggingface/diffusers/releases/tag/v0.3.0)

- Fixes previous bugs from versions 0.2.2 and 0.2.3
- Add support for gradient checkpointing
- Add [Google Colab notebook ](https://colab.research.google.com/drive/16m1HA3guhvP-uJRGMhD0gvuaMAUiiqUc?authuser=1#scrollTo=jKd5c5oUBmNP)