[2/n][Adjoint Module] Fix source deposition across chunk boundaries

[smartalecH]

Stack

Each PR is based on the one above it, so every diff here is just its own delta.

	PR
1	#3277	Fix MaterialGrid design-variable attribution (#1984)
2	#3290	Fix source deposition across chunk boundaries ← you are here
3	#3292	Fix adjoint source weighting on a symmetry plane (#3291)
4	#3280	Enable proper JAX support in the adjoint module
5	#3281	Differentiable angular-spectrum propagation
6	#3283	Adjoint gradients w.r.t. source amplitudes
7	#3284	Launch a mode through a stratified stack
8	#3285	Adjoint gradients w.r.t. geometry (centre and size)
9	#3286	Register a material grid's susceptibilities
10	#3287	Evaluate the dispersive adjoint at the discrete lineshape
11	#3293	Subpixel-smooth dispersive materials and their shape derivative
12	#3294	Scalable Pade extrapolation for DFT monitors (#3217)
13	#3295	3D grating coupler example: fiber, grating and mirror

Closes #1051

When users specify a geometry center, it shifts the locations of the sources relative to the chunk pattern (as expected). One bug that was fixed long ago, was when sources were accidentally placed on both chunks after a shift. Another issue (which is resolved here) is when a source shifts, but not all of it is properly shifted due to the way meep handles "ownership."

More specifically, when a zero-thickness source lies on a chunk boundary, Meep must interpolate it onto grid points on both sides. Previously, each chunk clipped that interpolation to its locally owned grid points. With a shifted simulation center, this could omit part of the source or store it on the wrong chunk, making results depend on the chunk layout.

The fix computes the complete source interpolation first, then moves each resulting source sample to the chunk that owns the corresponding field update. This makes source deposition independent of domain decomposition.

[smartalecH]

@lxvm can you try this out?

[lxvm]

Sure! Is there an issue I can reference to understand what this should fix?

[smartalecH]

@lxvm this should resolve the issue you were referencing in #3264 (i.e. the value of your objective function depends on the chunk layout)

[lxvm]

@smartalecH this pr fixes the consistency in my simulations w.r.t. MPI parallelization when I use geometry_center. Many thanks!

[smartalecH]

@stevengj note that I stacked this on #3277 (to make it easier for Lorenzo to run the full patch suite) so if we approve, we need to merge that one first and rebase here.

[smartalecH]

Closing in favour of #3298, which is the same branch and the same content, opened from this repository rather than from a fork.

The reason is purely mechanical: this series is a stack of 13 pull requests, each based on the one below it so that every diff is only its own delta. GitHub's native stacked pull requests cannot include a PR opened from a fork, and four of the thirteen — including this one — were. Rather than leave the series half-stacked, all four have been reopened with their branches hosted here so the whole chain can be linked.

Nothing about the change itself has been revised, and the description is carried over verbatim. The discussion and reviews on this PR stay readable here; please continue on #3298.