Possibly replace missing with NaN in convert functions

[biona001]

Currently convert_gt, convert_ht, and convert_ds represent missing data as missing. This causes the output genotype matrix to be of type Array{Union{Missing, T}, 2}. If we use NaN to represent missing instead, the output matrix type will be Array{Float64,2}, so BLAS will work by default. Is there any opinion on this?

It comes down to the following:

• NaN only works for floating-point numbers (i.e. no more integer matrices)
• NaN can be confusing because it can mean invalid math operations (e.g. division by 0) or the absence of a value.
• We lose functions like skipmissing