X-mlps

Improve the 1D and 2D shift operators. Specifically:

- Enable the 1D shift operator to operate on tensors of ndim >= 2.
- Enable the 2D shift operator to operate on tensors of ndim >= 3. Additionally, remove the height and width arguments to `create_shift2d_op` and assume the input data has the shape `(..., height, width, dim)`.

Move to using the Poetry package manager. Additionally, rename `SampleDropout` to `DropPath` and support per-sample dropout as implemented in timm.

Implement two 1D shift operations. In contrast to `create_shift2d_op`, these operations are useable with sequence data. These operations include:

- `create_shift1d_op`
- `create_multishift1d_op`

Additionally, update the README and add an RF/ML example to showcase the 1D shift operations.

- Implement two FNet inspired blocks.
- Split `_x_mlps.py` into several modules to reduce file size.
- Update README.md

Initial release. Create core MLP architecture and implement four concrete architectural blocks:

- MLP-Mixer
- ResMLP
- gMLP
- S^2-MLP