From the VCF designation https://samtools.github.io/hts-specs/VCFv4.2.pdf

Each row represents a new variant but because multiple reads are used to call variants it no longer has the information at the read level. It does however have positional information so you can likely use that to look in you BAM/SAM file for reads that span multiple variants. Some psuedocode might be:

for row in SAM:
for row in VCF:
    if SAM.start < VCF.position { Left = VCF.row }
    if SAM.end > VCF.position { Right = VCF.row }
done
if Left & Right { contains2 = True }
Spanning[row,] = [contains2, Left, Right]

Also, it shouldn't be a problem with the phasing because a single read is physically incapable of having sequence from both genomes of a diploid organisms since each read is from a single DNA molecule.