Adjoint TO filters not properly accounting for boundary conditions

[smartalecH]

As discussed, the current adjoint TO filters are not properly accounting for boundary conditions. In theory, we should only need a single pixel around the boundary. The current codebase does an edge-based pad on all dimensions:

meep/python/adjoint/filters.py

Line 83 in ee5324c

x = _edge_pad(x, ((kx, kx), (ky, ky)))

Maybe we need an even larger pad? (It's already padded twice what it needs to be to avoid overwriting values due to the circular convolution...)

(cc @oskooi )