alt_alleles dimension is not picked up when max_number is zero

[tomwhite]

Found while working on sgkit-dev/vcztools#198

Note that the Number=A becomes Number=1 in the following conversion:

$ bcftools view --header-only tests/data/vcf/1kg_2020_chr20_annotations.bcf | grep HWE_EUR_unrel
##INFO=<ID=HWE_EUR_unrel,Number=A,Type=Float,Description="HWE test in EUR (2504 unrelated samples) (PMID:15789306); 1=good, 0=bad">
$ vcf2zarr convert --force tests/data/vcf/1kg_2020_chr20_annotations.bcf 1kg_2020_chr20_annotations.vcz
$ vcztools view --header-only 1kg_2020_chr20_annotations.vcz | grep HWE_EUR_unrel
##INFO=<ID=HWE_EUR_unrel,Number=1,Type=Float,Description="HWE test in EUR (2504 unrelated samples) (PMID:15789306); 1=good, 0=bad">

This is similar on the surface to #353, but there's a deeper problem to do with handling the case where the VCF header declares Number=A, but the max_number found when writing the ICF is 0.