v5.1.6

Fixed

• Batch LMM memory preflight now propagates n_cvt to estimate_lmm_memory at both call sites (PipelineRunner._run_inner batch branch and run_lmm_association_numpy). Previously these passed only (n_samples, n_snps), silently defaulting n_cvt=1, so multi-covariate runs could pass the preflight and then OOM at the real Uab_batch / Iab_batch allocations (which scale with n_cvt). The streaming branch was already correct.

See CHANGELOG.md for full history.