Add more general int64 conversion #4
Description
Int64 arrays currently require either fitting into 32bits (after offset subtraction) or conversion to float64. Instead, we could more generally split the upper and lower 32bit values and compress them separately.