Investigate performance implications of using Java serialization for genotypes

[heuermh]

Pull request #52 uses JavaSerializer for htsjdk.variant.variantcontext.LazyGenotypesContext to workaround issues with serializing transient fields.

Java binary serialization is demonstratively worse than Kryo, in both read/write speed and size in memory, so we should follow up with an investigation of how much this matters for performance in Disq for genotypes. Other options might include: modify LazyGenotypesContext so that it can be serialized by Kryo, write a custom Kryo serializer for LazyGenotypesContext, etc.

[heuermh]

Worth revisiting this to see if still an issue with the latest htsjdk.