v5.4.5

Fixed

Single-point lambda grids are rejected at the config boundary. LmmConfig and PipelineConfig both accepted n_grid=1, though a one-point grid has no bracket to refine. The C kernel rejects it in validate_batch_params, but only once the run reaches the kernel — after kinship and eigendecomposition have been paid for — and the NumPy fallback has no such check, so it silently returned lambda = l_min for every SNP instead of its optimum.

Both configs now raise ValueError on construction (n_grid must be >= 2), which also covers the LOCO branch, where n_grid is forwarded to run_lmm_loco without an LmmConfig ever being built.

Full changelog: v5.4.4...v5.4.5