Clarification of empty vector representation

[pd3]

as triggered by samtools/bcftools#1622 and #593

State explicitly that missing vectors (.) in fields of variable length are in BCF represented as MISSING,END_OF_VECTOR, i.e. the leading MISSING value must be always present.

[hts-specs-bot]

Changed PDFs as of 9a5fdb0: VCFv4.3 (diff). This link will expire in 30 days

[jmarshall]

#593 makes a case that [END_OF_VECTOR, END_OF_VECTOR] would also be a reasonable encoding of this situation.

What is the reason for requiring that “[e]mpty vectors (i.e. vectors with no data available)” be represented as [MISSING, END_OF_VECTOR] (which is seemingly less empty!)?

[jkbonfield]

Fixes #593