Improve jvec/jtvec

[prisae]

e4d6ac8 introduced jvec and jtvec. There are a couple of things that need improvements down the road. An issue to keep track of it:

• Restructure the internal layout of gradient/_bcompute/_get_rfield/_jvec/_jtvec
• Expand documentation, adding the derivation for jvec and jtvec in more detail
• Optimize the entire gradient-chain for speed
• Add an example to the gallery using SimPEG(emg3d)

Also, testing should be improved:

• Add adjoint tests and derivative tests using discretize
  • src-rec
  • mapping
  • interpolation
  • jvec/jtvec (overall)

More work is required to make it possible to have automatic gridding ('gridding' != 'same').

• SimPEG@emg3d-solver needs adjustments to work with different inversion and computational grids
• jvec: vector should probably be a Model (needs adjustments in SimPEG@emg3d-solver)
• jtvec: vector should probably be a DataArray (needs adjustments in SimPEG@emg3d-solver) (both is already possible, ndarray or DataArray)
• Check interpolations; maps.interp_edges_to_vol_averages; grid.get_edge_inner_product_deriv
• Automatic gridding SimPEG@emg3d-solver: Which model is used when to create grids? Needs work

Current relevant branches:

• emg3d:
  • https://github.com/emsig/emg3d/tree/dev-simpeg-autogridding
  • https://github.com/emsig/emg3d/tree/gradient-variants
  • main or emg3d>=v1.5.0
• SimPEG:
  • https://github.com/simpeg/simpeg/tree/emg3d-solver
  • https://github.com/simpeg/simpeg/tree/emg3d-solver-new