Generalize `convert_probability_to_call` to multiallelic and polyploid cases

[timothymillar]

This is currently only implemented for the diploid biallelic case. Generalizing it to the multiallelic case would require that the number of possible alleles at each variant is known. See relevant discussion in #1006 about counting alleles at each variant. We can preserve backwards compatibility by assuming the diploid biallelic case when the relevant dimensions are absent (validating that the size of the probabilities array is correct).