E3nn

What's Changed
* PT2 compile with Legacy code by mitkotak in https://github.com/e3nn/e3nn/pull/455
* Release fix by mitkotak in https://github.com/e3nn/e3nn/pull/469
* update deprecated matplotlib option by eszter137 in https://github.com/e3nn/e3nn/pull/475
* Fix docs build by mitkotak in https://github.com/e3nn/e3nn/pull/476
* Updating PyPI token in release process by mitkotak in https://github.com/e3nn/e3nn/pull/478
* updated citations format to make twine happy by mitkotak in https://github.com/e3nn/e3nn/pull/479
* Bump python version to only 3.11 in tests by mitkotak in https://github.com/e3nn/e3nn/pull/487
* Small change to pyproject.toml to fix pypi docs by rmcconke in https://github.com/e3nn/e3nn/pull/488
* PyPI release 0.5.4 by mitkotak in https://github.com/e3nn/e3nn/pull/489

New Contributors
* eszter137 made their first contribution in https://github.com/e3nn/e3nn/pull/475
* rmcconke made their first contribution in https://github.com/e3nn/e3nn/pull/488

**Full Changelog**: https://github.com/e3nn/e3nn/compare/0.5.2...0.5.4

Added

- `o3.experimental.FullTensorProductv2 | ElementwiseTensorProductv2` for compatibility with `torch.compile(..., fulgraph=True)`
- enable pip caching in CI
- Optional scalar bias term in _batchnorm.py

Changed

- refactor to use pyproject.toml for packaging
- refactor gh community files
- move pylint, coverage and flake8 configuration to pyproject.toml

Fixed

- Fix TorchScript warning "doesn't support instance-level annotations" (437)

Added
- L=12 spherical harmonics

Fixed
- `TensorProduct.visualize` now works even if the TP is on the GPU.
- Github actions only trigger a push to coveralls if the corresponding token is set in github secrets.
- Batchnorm

Added
- Sparse Voxel Convolution
- Clebsch-Gordan coefficients are computed via a change of basis from the complex to real basis. (see https://github.com/e3nn/e3nn/pull/341)
- `o3`, `nn` and `io` are accessible through `e3nn`. For instance `e3nn.o3.rand_axis_angle`.

Changed
- Since now the code is no more tested against `torch==1.8.0`, only tested against `torch>=1.10.0`

Fixed
- `wigner_3j` now _always_ returns a contiguous copy regardless of dtype or device

Fixed
- Remove `CartesianTensor._rtp`. Instead recompute the `ReducedTensorProduct` everytime. The user can save the `ReducedTensorProduct` to avoid creating it each time.
- `*equivariance_error` no longer keeps around unneeded autograd graphs
- `CartesianTensor` builds `ReducedTensorProduct` with correct device/dtype when called without one

Added
- Created module for reflected imports allowing for nice syntax for creating `irreps`, e.g. `from e3nn.o3.irreps import l3o same as Irreps("o3")`
- Add `uvu<v` mode for `TensorProduct`. Compute only the upper triangular part of the `uv` terms.
- (beta) `TensorSquare`. computes `x \otimes x` and decompose it.
- `*equivariance_error` now tell you which arguments had which error

Changed
- Give up the support of python 3.6, set `python_requires='>=3.7'` in setup
- Optimize a little bit `ReducedTensorProduct`: solve linear system only once per irrep instead of 2L+1 times.
- Do not scale line width by `path_weight` in `TensorProduct.visualize`
- `*equivariance_error` now transforms its inputs in float64 by default, regardless of the dtype used for the calculation itself

Fixed
- `ReducedTensorProduct`: replace QR decomposition by `orthonormalize` the projector `X.T X`.
This keeps `ReducedTensorProduct` deterministic because the projectors and `orthonormalize` are both deterministic.
The output of `orthonormalize` apears also to be highly sparse (luckily).