[profsergiocosta]

Compared with TerraME 2.0.1 runs of the ca package (terrame-goldens v0.1.0), three models differed:

• GameOfLife (vector and raster) did not wrap around the edges; TerraME's Life uses createNeighborhood{wrap = true}. New wrap parameter, default True.
• FireModelRaster passed no neighborhood to focal_sum_mask, which counts the eight Moore neighbors, so the fire spread diagonally; it now uses self.dirs (Von Neumann), as TerraME and the vector version.
• SolidDiffusion drew only among atoms and could move a vacancy several times per step. It now follows TerraME: each vacancy draws one of all its neighbors, reads the state at the start of the step and moves at most once; cells in TerraME's order, on NumPy arrays.

With these fixes, the 14 deterministic cases match TerraME cell by cell at every step, and SolidDiffusion mixes as in TerraME (26.5 vs 27 atoms past the middle column at t = 400; 35.9 before).

tests/test_goldens.py (marker golden) runs the comparisons from an extract of the goldens in tests/goldens (with SHA-256 provenance and the script that makes it): grids cell by cell, ensembles by Kolmogorov-Smirnov and mean tests. About a minute by default; DISSMODEL_GOLDEN_FULL=1 for every step and the slow ensembles.