Volume preserving feedforward neural network (vpffnn)

[benedict-96]

Added volume-preserving feedforward neural network (vpffnn). The corresponding layers essentially perform:

$$\psi: x \mapsto \sigma(Lx + b),$$

where $L$ is a lower-triangular layer (the corresponding version for the upper triangular layer also exists). By construction $\mathrm{det}(\nabla\psi) = 1$ and the layer is therefore volume-preserving. What will be used in the end is the neural network architecture VolumePreservingFeedForward.

A big part of this pr are docs and various tikz pictures explaining the constructions made here. There are also tests.

[codecov]

Codecov Report

Attention: Patch coverage is 62.28814% with 178 lines in your changes are missing coverage. Please review.

Project coverage is 62.53%. Comparing base (908bc21) to head (c2d7eb6).

Files	Patch %	Lines
src/arrays/triangular.jl	18.57%	57 Missing ⚠️
src/architectures/volume_preserving_feedforward.jl	2.85%	34 Missing ⚠️
src/kernels/kernel_ad_routines/mat_tensor_mul.jl	78.70%	33 Missing ⚠️
src/architectures/neural_network_integrator.jl	0.00%	26 Missing ⚠️
src/kernels/mat_tensor_mul.jl	90.00%	8 Missing ⚠️
src/arrays/lower_triangular.jl	78.12%	7 Missing ⚠️
src/arrays/upper_triangular.jl	80.55%	7 Missing ⚠️
src/layers/volume_preserving_feedforward.jl	84.21%	6 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #128      +/-   ##
==========================================
- Coverage   62.58%   62.53%   -0.05%     
==========================================
  Files         105      111       +6     
  Lines        2750     3222     +472     
==========================================
+ Hits         1721     2015     +294     
- Misses       1029     1207     +178     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.