concordance: support aliasing imputed sample IDs to truth-VCF IDs

[tfenne]

Summary

GLIMPSE2_concordance previously matched samples between the imputed VCF and
the validation VCF by exact sample-ID equality. This blocks a common
evaluation workflow: testing the same biological sample (e.g. NA12878)
imputed under many conditions (e.g. multiple downsamplings, different GLIMPSE2_phase settings, etc.) against a single truth row.

This PR extends the existing --samples file with an optional second
whitespace-separated column per row giving the corresponding sample ID in the
validation VCF. The second column is optional on a per-row basis, so a single
file may freely mix aliased and non-aliased rows. Multiple imputed rows may
alias the same validation row.

# samples.txt -- first column = imputed ID, optional second column = validation ID
NA12878.0_5x  NA12878
NA12878.1x    NA12878
NA12878.2x    NA12878
HG00096
HG00097

Implementation notes

• The validation-side header is subset to only the K unique validation
  IDs needed (K <= N imputed rows). A length-N imputed_to_truth_col[i]
  array maps each imputed row to its column in the validation subset.
  This preserves the existing per-record decode cost in the common case
  (many imputed reps -> one truth sample): the validation side still
  decodes K columns per record, not the full validation file.
• Per-sample output rows are now labelled with the imputed-side ID so each
  replicate appears as its own row.
• A defensive per-record check now fails with a clear error if the
  validation file's per-sample PL/GT stride disagrees with the
  imputed-side ploidy at a site (reachable when the user aliases samples
  of differing ploidy).
• Documentation (docs/docs/documentation/concordance.md) updated with a
  "Sample aliasing" section and an expanded --samples description.

Backwards compatibility

• 1-column --samples files behave exactly as before.
• No---samples runs produce the same per-sample numbers as before. The
  per-sample output row order changes from std::unordered_set
  iteration order (i.e. hash-bucket order, libstdc++-version-dependent)
  to imputed-VCF header order. Strictly an improvement in determinism;
  flagging since it affects diffs of *.error.spl.txt.gz and
  *.rsquare.spl.txt.gz.

Testing

The repo has no unit-test framework, and the tutorial pipeline doesn't
exercise sample subsetting, so the new path was tested manually:

• The tutorial concordance step (tutorial/step7_script_concordance.sh)
  was re-run after each commit; all five output files
  (output.error.{spl,grp,cal}.txt.gz, output.rsquare.{grp,spl}.txt.gz)
  are bit-identical to a pre-change baseline.
• The aliasing path was exercised by running with a renamed copy of the
  imputed VCF and a 2-column --samples file aliasing the renamed ID
  back to the original truth ID; per-sample numbers matched the baseline
  with the imputed-side ID labelling the row.
• The new ploidy-mismatch check was verified by forcing a haploid
  validation VCF (via bcftools +fixploidy -f 1) against the original
  diploid imputed file; the check fails with a clear chrom:pos error
  instead of producing corrupted output.