WaveletOperator fixes for odd-sized inputs and 1D (VectorGeometry) - #1627

Describe changes

• In adjoint, after the PyWavelet reconstruction, the array is always cut to correct size in case the domain_geometry is of odd size. The tuple of slices org_size could be created during initialization as well.
  • Channels should work as is, even if they have odd length
• When initializing the range_geometry for 1D (VectorGeometry) data, the correct shape parameters are given.
  • Channels most likely will not work but I do not know if that behaviour is supported anyways.
• Cleaned some old comments and added few new ones, nothing major

Describe any testing you have performed

• No proper testing, just some of my own experiments.
• The WaveletOperator should have a proper tests built at some point because there are so many things that can go wrong. For example
  • Check that the dimensions of input and output are correct
  • Check that odd sized arrays behave correctly
  • Check that decomposing different axes works correctly

Link relevant issues

• All issues related to too big decomposition level are left to PyWavelets to handle. There should be a warning if very long filters are used with too small data / too big decomposition level, because this leads to nonsense coefficients.
• In addition to axes for choosing the dimensions the operator acts on, there should be string-based correlation parameter which would define the axes similar to GradientOperator

Checklist when you are ready to request a review

• I have no idea what I am doing