Simplify properties

[AntonOresten]

Closes #8

I did a little more than simplifying the properties, adding an atoms field to ProteinStructure, and removing the backbone field from ProteinChain. This is breaking and intended to be pushed with a version that easily serializes to HDF5.

Could use some more unit tests.

[bicycle1885]

This looks good to me. Are you going to add the serializer to this PR?

[AntonOresten]

Yes, here it is. I currently define ProteinChains.serialize and ProteinChains.deserialize which are meant work with collections of ProteinStructures, so there's currently no way to write single chains or structures. Is it worth having some ProteinDataset <: AbstractDict{String,ProteinStructure} type for easy indexing and unambiguous serialization IO, or are vectors good enough?

[bicycle1885]

I tried the serialization and deserialization pair, which works nicely. One practical thing we'll need is partial/sequential deserialization from a serialized data file because the file would be too large to deserialize in one go. I assume some interface like the following:

data = Deserializer("data.h5")

# Sequential deserialization
for structure in data
    # Do something
end

# Partial deserialization
for id in some_id_list
    structure = data[id]  # Or multiple ids if it is more efficient
    # Do something
end

This can be done in a separated pull request.

[AntonOresten]

Great idea with the lazy deserialization type. I might also try a corresponding "Serializer" with e.g. a Base.delete! method, as the current serialize mode keyword argument doesn't give a lot of control. Opening a PR soon!