fix: writeVcf restores '*' spanning deletion in multi-allele ALT (#65)

[jmg421]

Summary

Partially fixes #65.

writeVcf() now correctly writes * (spanning deletion) alleles in multi-allele ALT fields instead of producing empty alleles that make the VCF malformed.

Problem

As reported in #65, writing a VCF with spanning deletions produced output like:

chr1  10150  .  CTA  CA,,TTA  ...

instead of:

chr1  10150  .  CTA  CA,*,TTA  ...

The empty allele caused IGV/htsjdk to reject the file with:

empty alleles are not permitted in VCF records

Root Cause

readVcf() converts * to empty strings at parse time (in .formatALT()), losing the distinction between * (spanning deletion) and . (no allele). At write time, these empty strings were collapsed into empty fields.

Fix

At write time (.makeVcfMatrix()), before collapsing ALT alleles with unstrsplit, restore * for empty strings that appear in multi-allele list elements (length > 1). An empty string alongside other alleles is unambiguously a spanning deletion per VCF spec. For ExpandedVCF objects (scalar ALT per row), empty strings also become * since expand() doesn't produce monomorphic rows.

Limitations

The ambiguous single-allele case (a sole * vs .) cannot be fully resolved at write time because the C-level parser erases the distinction. This PR conservatively writes . for that case. A complete fix would require changes to the C parsing layer as @hpages discussed in the issue thread.

Testing

• All existing writeVcf tests pass
• Verified manually: CA,*,TTA is correctly round-tripped through read/write

Agentic tooling disclosure

This PR was produced with the assistance of Kiro CLI (Amazon's AI coding agent). All changes were reviewed and verified by the author.

[jmg421]

Superseded by #104, which fixes the issue at read time (preserving '' as a CharacterList element) rather than trying to reconstruct it at write time. The read-time fix is cleaner, handles all cases (including single '' alleles), and aligns with how structural variants are already handled.