backwards diffs for JopGradient on edges

[jkwashbourne-oss]

adds backwards one sided diffs on the edges where previously no derivatives were computed.

thinking more about it, perhaps this results in unexpected behavior, since the last two points in the dimension we are taking the derivatives will have the same value.

julia> nz,nx = 5,5
(5, 5)

julia> dz,dx = 25.0,25.0
(25.0, 25.0)

julia> G = JopGradient(JetSpace(Float32,nz,nx), (Float32(dz), Float32(dx)))
"Jet linear operator, (5, 5) → (5, 5, 2)"

julia> x = rand(Float32,nz,nx)
5×5 Matrix{Float32}:
 0.957345   0.534336   0.144943  0.424699  0.727155
 0.449017   0.886986   0.303386  0.235856  0.764744
 0.980932   0.391082   0.116915  0.202955  0.598075
 0.0956441  0.0800584  0.623844  0.639636  0.365211
 0.532148   0.491575   0.325524  0.058127  0.24094

julia> G*x
5×5×2 Array{Float32, 3}:
[:, :, 1] =
 -0.0203331   0.014106    0.00633772  -0.00755372   0.00150356
  0.0212766  -0.0198362  -0.00745887  -0.00131601  -0.00666677
 -0.0354115  -0.0124409   0.0202772    0.0174672   -0.00931453
  0.0174601   0.0164607  -0.0119328   -0.0232604   -0.00497085
  0.0174601   0.0164607  -0.0119328   -0.0232604   -0.00497085

[:, :, 2] =
 -0.0169204    -0.0155757    0.0111902     0.0120983    0.0120983
  0.0175188    -0.023344    -0.00270122    0.0211555    0.0211555
 -0.023594     -0.0109667    0.00344164    0.0158048    0.0158048
 -0.000623429   0.0217514    0.000631707  -0.010977    -0.010977
 -0.00162291   -0.00664205  -0.0106959     0.00731252   0.00731252