v0.104.0

• API: FFTEngine field methods now use components parameter instead of nb_components
  • real_space_field(name, components=()) - creates scalar field by default
  • fourier_space_field(name, components=()) - creates scalar field by default
  • register_real_space_field(name, components=()) - creates scalar field by default
  • register_fourier_space_field(name, components=()) - creates scalar field by default
  • This makes the API consistent with FieldCollection.real_field() and CartesianDecomposition.real_field()
• API: Scalar fields now have different shapes than unit component fields
  • Scalar field (components=()): .p.shape = (nx, ny), .s.shape = (1, nx, ny)
  • Unit component (components=(1,)): .p.shape = (1, nx, ny), .s.shape = (1, 1, nx, ny)
  • Use components=(1,) explicitly if you need the component dimension
• ENH: Added Shape_t overloads for all FFTEngine field registration methods
  • Supports arbitrary component shapes like (3,) for vectors or (3, 3) for tensors
  • Available for CPU, CUDA, and ROCm backends
• DOC: Updated fourier_derivative.py example for scalar field indexing