fix(heteroplasmy): address adversarial review findings

- DOWNSAMPLE_MATCH: include samples at/above T via float compare (was -gt on a rounded int, which dropped boundary samples). - MERGE_VCFS_HET: fall back to the single VCF when <2 samples so Haplocheck still runs on small pilots (bcftools merge needs >=2). - PARSE_HETEROPLASMY: drop '|| true' so hard bcftools failures surface (errorStrategy 'ignore' keeps step 3 alive). - MERGE_HETEROPLASMY: defensive split('-') on sample_id, matching MERGE_MTDNA_METRICS. - wiring: document why nat_vcfs uses a bare .collect(). Three review findings were verified as false positives and left as-is.