pecotmr fixes

tests/fixtures/mash/expected_manifest.tsv

@@ -10,7 +10,7 @@ cov.canonical.EE.rds	tolerant	1e-6	1e-8	mash_covariance.R canonical component li
 cov.pca.EE.rds	tolerant	1e-6	1e-8	mash_covariance.R pca component list (4 matrices, npc 3, EE); deterministic.
 cov.flash.EE.rds	tolerant	1e-6	1e-8	mash_covariance.R flash component list (EE); flashier self-seeds — empty on the weak toy input (0 factors), a valid regression baseline.
 cov.flash_nonneg.EE.rds	tolerant	1e-6	1e-8	mash_covariance.R flash_nonneg component list (EE); empty on weak toy input (0 factors).
-mash_sumstats.region1.rds	tolerant	1e-6	1e-8	QtlSumStats S4 (40 variants x 2 conditions) from mash_sumstats_construct.R over tensorqtl z-score TSVs; deterministic, no embedded paths.
+mash_sumstats.region1.rds	tolerant	1e-6	1e-8	QtlSumStats S4 (40 variants x 2 conditions) from mash_sumstats_construct.R over tensorqtl z-score TSVs; deterministic, no embedded paths. Re-baselined for pecotmr 0.7.8 (#591 class cleanup): the genome build moved off the SumStatsBase `genome` slot into seqinfo(), so this fixture now records genome=GRCh38; values are unchanged (old-vs-new all.equal 1e-6 with the build neutralized).
 mash_model.EE.rds	tolerant	1e-6	1e-8	mash_fit.R fitted model list(mash_model, vhat_file, prior_file) over the mash_model_chain (vhat=simple -> prior cov_ed canonical+pca npc 3 -> fit), EE; reproducible. vhat_file/prior_file embed the chain tmp paths -> compared with normalize_paths=True (rds_compare normPaths basenames the character leaves).
 posterior.EE.rds	tolerant	1e-6	1e-8	mash_posterior.R posterior list (PosteriorMean/SD/lfdr/NegativeProb/lfsr/PosteriorCov; 17 variants x 8 conditions) on strong.b/strong.s given the fitted model, EE; deterministic, no embedded paths.
 posterior_contrast.rds	tolerant	1e-6	1e-8	mash_posterior_contrast.R contrast data.frame (17 variants x 108 contrasts) from the mash_posterior fixture, cells ALL..Oli; deterministic. row.names are stable variant ids (mash::mash::varN), no embedded paths.

tests/fixtures/qtl_association_postprocessing/expected/expected_manifest.tsv

@@ -1,5 +1,5 @@
 # expected-output regression manifest — qtl_association_postprocessing
 # columns: output <tab> mode <tab> rtol <tab> atol <tab> notes
-qap.rds	tolerant	1e-6	1e-8	qtl_association_postprocessing.R enriched QtlSumStats S4 (10 genes) via pecotmr::qtlAssociationPostprocess on the committed cis-QTL fixture; deterministic, no embedded paths.
+qap.rds	tolerant	1e-6	1e-8	qtl_association_postprocessing.R enriched QtlSumStats S4 (10 genes) via pecotmr::qtlAssociationPostprocess on the committed cis-QTL fixture; deterministic, no embedded paths. Re-baselined for pecotmr 0.7.8 (#591 class cleanup): the genome build moved off the SumStatsBase `genome` slot into seqinfo(), so this fixture now records genome=hg38; values are unchanged (old-vs-new all.equal 1e-6 with the build neutralized).
 qap.cis_regional.fdr.tsv.gz	tolerant	1e-6	1e-8	Per-gene enriched regional export (input regional + hierarchical multiple-testing correction columns); deterministic, no embedded paths.
 qap.summary.tsv	tolerant	1e-6	1e-8	Per-method significant-event / significant-QTL counts summary at the default 0.05 FDR (all zero on the toy); deterministic, no embedded paths.

tests/fixtures/rss_analysis/expected/expected_manifest.tsv

@@ -1,4 +1,4 @@
 # expected-output regression manifest — rss_analysis (notebook tier)
 # columns: output <tab> mode <tab> rtol <tab> atol <tab> notes
-gwas_sumstats.rds	tolerant	1e-6	1e-8	generate_gwas_sumstats cell -> gwas_sumstats_construct.R (GwasSumStats for AD_Bellenguez_2022 chr22:49355984-50799822 over the toy chr22 GWAS + ld_reference panel). Deterministic QC/reshaping (no RNG); bit-identical run-to-run. normalize_paths: the ldSketch slot embeds the absolute ld_reference LD-panel path (resolved from --ld-meta at the checkout location), basename-normalized by rds_compare.R.
+gwas_sumstats.rds	tolerant	1e-6	1e-8	generate_gwas_sumstats cell -> gwas_sumstats_construct.R (GwasSumStats for AD_Bellenguez_2022 chr22:49355984-50799822 over the toy chr22 GWAS + ld_reference panel). Deterministic QC/reshaping (no RNG); bit-identical run-to-run. normalize_paths: the ldSketch slot embeds the absolute ld_reference LD-panel path (resolved from --ld-meta at the checkout location), basename-normalized by rds_compare.R. Re-baselined for pecotmr 0.7.8 (#591 class cleanup): the genome build moved off the SumStatsBase `genome` slot into seqinfo(), so this fixture now records genome=GRCh38; values are unchanged (old-vs-new all.equal 1e-6 with the build neutralized).
 gwas_finemap.rds	tolerant	1e-6	1e-8	gwas_fine_mapping cell -> fine_mapping.R (GwasFineMappingResult, susie). The notebook cell forwards no --seed but susie_rss is deterministic for this fixed input+init: bit-identical across two full pipeline reruns on this machine. Same embedded ld_reference path -> normalize_paths. The x86/ARM CI matrix may need a looser rtol (calibrate from the first CI run), mirroring fine_mapping/expected_manifest.tsv.

tests/fixtures/twas/expected_manifest.tsv

@@ -1,4 +1,4 @@
 # expected-output regression manifest — twas
 # columns: output <tab> mode <tab> rtol <tab> atol <tab> notes
-gwas_sumstats.chr22.rds	tolerant	1e-6	1e-8	GwasSumStats S4 (one chr22 LD block) from gwas_sumstats_construct.R; deterministic. Compared with normalize_paths=True — the S4 embeds the machine-specific LD-reference path.
+gwas_sumstats.chr22.rds	tolerant	1e-6	1e-8	GwasSumStats S4 (one chr22 LD block) from gwas_sumstats_construct.R; deterministic. Compared with normalize_paths=True — the S4 embeds the machine-specific LD-reference path. Re-baselined for pecotmr 0.7.8 (#591 class cleanup): the genome build moved off the SumStatsBase `genome` slot into seqinfo(), so this fixture now records genome=GRCh38; values are unchanged (old-vs-new all.equal 1e-6 with the build neutralized).
 twas.chr22.rds	tolerant	1e-6	1e-8	TWAS-Z GRanges from twas.R (causalInferencePipeline) over the committed S4 TwasWeights + a built GwasSumStats; deterministic, no embedded paths.