[JerryIshihara]

New synthetic test fixture and infrastructure:

• Added generate_synthetic_dataset.py to create a deterministic synthetic three-program cNMF dataset, including known ground-truth usages and spectra, and outputs in both AnnData and TSV formats.
• Added generate_original_results.py to automate running cNMF on the synthetic dataset, capturing outputs at each stage, tracking NMF iteration counts, and generating a manifest for reproducibility.
• Added fixture manifest (manifest.yaml) and all necessary input and reference output files, including genes.txt, synthetic_counts.tsv, true_spectra.tsv, and true_usages.tsv, to enable deterministic and transparent regression testing. [1] [2] [3] [4] [5] [6] [7]

QC pipeline improvements:

• The run_qc function in src/cnmf/qc.py now accepts and processes normalized counts and refit usages, computes R², SSE, and TSS metrics, and includes them in the QC annotation output. [1] [2]
• The cNMF consensus function is updated to pass normalized counts and refit usages through to the QC step, ensuring these new metrics are available for downstream analysis and validation. [1] [2]