ModelScaleGPU: a Levenberg-Marquardt fit whose J^T J has underflowed to 0 ends at its best so far, as on the CPU

Since fdafc009b J^T J and J^T r are summed in float. A fit that walks to a nonsensical scale (k_overall < 0,
B of thousands of A^2, as fits of an unrelated model can in the solvent grid) makes |F| k_aniso small enough
that every square underflows to exactly 0 in float, while J^T r keeps a residue near 1e-36. JordanSolve then
threw "Trying to reverse singular matrix. Column 0 is zeroed." out of ValidateAgainstModel
(ModelValidation_CCModelUnrelatedModelConfirmsNothing). In double the squares stayed near 1e-72 and the
solve went through. LevMarFit in ModelScaling.cpp already stops at alpha[0] == 0 for the same case in
double; the GPU run now does the same check before the solve.

Only fits that would have thrown are affected; no kernel changes, deterministic as before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HVeDQqN9qRD57vnShzTC37

image_analysis/structure_refinement/ModelScaleGPU.cu

@@ -373,6 +373,14 @@ struct ModelScaleGPU::LevMarRun {
             done = true;
             return;
         }
+        // As LevMarFit (ModelScaling.cpp): a fit that has walked to where k_overall exp(-B s^2) |F| underflows at
+        // every reflection has no step left - the squares in alpha are all 0 while beta keeps a residue, and the
+        // solve throws on that. It ends here, at its best so far. J^T J is summed in float, so its squares
+        // underflow once |F| k_aniso is below about 1e-22 everywhere, still far from any scale a fit can end at.
+        if (alpha[0] == 0) {
+            done = true;
+            return;
+        }
         const size_t na = beta.size();
         std::vector<double> temp_alpha = alpha;
         for (size_t j = 0; j < na; j++)