Defer the eager scipy.sparse and scipy.spatial imports (~280 ms)

`import autoarray` drops from 464.4 ms to 183.7 ms (medians of 15 runs,
Python 3.13, dev extras) — a 281 ms saving.

The filed task named `derivative_util.py:30`'s module-scope
`from scipy.sparse import csr_matrix` as the target, on the evidence that
`scipy.sparse` costs ~0.11 s of every import. Deferring it changed nothing,
because `scipy.sparse` was never being imported from there:

    autoarray/__init__.py:80
      -> inversion/mesh/mesh_geometry/delaunay.py:2   import scipy.spatial
        -> scipy/spatial/__init__.py:111              from ._kdtree import *
          -> scipy/spatial/_kdtree.py:4               from ._ckdtree import cKDTree

`scipy.spatial` (134 ms) pulls `scipy.sparse` (154 ms) in transitively, so the
csr_matrix import was riding on a subtree that was already paid for. Deferring
`scipy.spatial` as well is what actually removes both, and is required to meet
the task's own acceptance criterion.

Changes:
- `inversion/mesh/mesh_geometry/delaunay.py` — drop the module-scope
  `import scipy.spatial`. Two of its three use sites already had local
  imports; only `voronoi_neighbors_from` needed one added, so this finishes a
  deferral that had been started and left half-done.
- `operators/derivative_util.py`, `operators/coarse_interp_util.py` — move
  `from scipy.sparse import csr_matrix` into the four functions that use it.
  No longer load-bearing for the import time on its own, but it keeps
  `scipy.sparse` off the import path independently of what `scipy.spatial`
  happens to pull in.

Every use site is inside a function, so a plain local import suffices — no
module-level cache as in `transformer.py:_load_nufftax()`, which needed one
only because unpickled instances in multiprocessing workers never re-run
`__init__`. Function-local imports run on every call and hit `sys.modules`
after the first.

Verified: `python -X importtime -c "import autoarray"` shows neither
`scipy.sparse` nor `scipy.spatial`. Suites green — autoarray 1179 passed,
autogalaxy 1103 passed / 1 skipped, autolens 532 passed / 1 skipped.

Author: claude

autoarray/inversion/mesh/mesh_geometry/delaunay.py

@@ -1,5 +1,4 @@
 import numpy as np
-import scipy.spatial
 from typing import Tuple
 
 from autonerves import cached_property
@@ -147,6 +146,7 @@ def neighbors(self) -> Neighbors:
         The neighbors of a Voronoi mesh are computed using the `ridge_points` attribute of the scipy `Voronoi`
         object, as described in the method `voronoi_neighbors_from`.
         """
+        import scipy.spatial
 
         delaunay = scipy.spatial.Delaunay(self.mesh_grid_xy)
 

autoarray/operators/coarse_interp_util.py

@@ -17,7 +17,6 @@
 """
 
 import numpy as np
-from scipy.sparse import csr_matrix
 
 from autoarray import exc
 from autoarray import numba_util
@@ -242,6 +241,8 @@ def coarse_interp_matrix_from(
     A ``scipy.sparse.csr_matrix`` of shape
     [n_unmasked_fine_pixels, n_unmasked_coarse_pixels].
     """
+    from scipy.sparse import csr_matrix
+
     mask_itp_box = np.asarray(mask_itp_box)
     if np.count_nonzero(~mask_itp_box) == 0:
         raise exc.MeshException(

autoarray/operators/derivative_util.py

@@ -27,7 +27,6 @@
 """
 
 import numpy as np
-from scipy.sparse import csr_matrix
 
 from autoarray import exc
 from autoarray import numba_util
@@ -297,9 +296,11 @@ def derivative_1st_operators_from(mask, pixel_scale: float = 1.0):
     -------
     The (Hy, Hx) operators as ``scipy.sparse.csr_matrix``.
     """
+    from scipy.sparse import csr_matrix
+
     mask, diff_types = _diff_types_of_cleaned_mask_from(mask)
-    rows_hx, cols_hx, data_hx, rows_hy, cols_hy, data_hy = (
-        derivative_1st_triplets_from(mask, diff_types, dpix=pixel_scale)
+    rows_hx, cols_hx, data_hx, rows_hy, cols_hy, data_hy = derivative_1st_triplets_from(
+        mask, diff_types, dpix=pixel_scale
     )
 
     n_unmasked = np.count_nonzero(~mask)
@@ -443,6 +444,8 @@ def derivative_2nd_operators_from(mask, pixel_scale: float = 1.0):
     -------
     The (Hyy, Hxx) operators as ``scipy.sparse.csr_matrix``.
     """
+    from scipy.sparse import csr_matrix
+
     mask, diff_types = _diff_types_of_cleaned_mask_from(mask)
     rows_hxx, cols_hxx, data_hxx, rows_hyy, cols_hyy, data_hyy = (
         derivative_2nd_triplets_from(mask, diff_types, dpix=pixel_scale)
@@ -574,6 +577,8 @@ def forward_difference_operators_from(
     -------
     The (Hy, Hx) operators as ``scipy.sparse.csr_matrix``.
     """
+    from scipy.sparse import csr_matrix
+
     if max_order not in (1, 2, 3, 4):
         raise ValueError(f"max_order must be in 1..4, got {max_order}")