Forward Exponax Release

[Ceyron]

Updates to Exponax 0.1.0 and forward all API changes.

Since Exponax changed some internals, this can affect some benchmarks. The most important changes are:

1. Change in the metric system (normalized metrics now normalized per channel and not over all channels)
2. Convection Nonlinearity changed to a non-conservative form by default ($b_1 \frac{1}{2} \nabla \cdot (u \otimes u) \rightarrow b_1 (u \cdot \nabla) u$), but the old "conservative" version is still available. I decided to keep it as the default in APEBench via setting the conservative=True for all steppers using convection. This is to ensure legacy compatibility
3. For efficiency, Substeppers now operate in Fourier space. Qualitatively, this should not alter produced trajectories but can cause numeric differences, especially for chaotic problems (KS, etc.). This can slightly alter benchmark runs numeric results. To avoid the influence of the concrete dataset, consider rerunning benchmark runs with different train and test seeds.