Slight redesign of the I/O system and binary I/O for Array

[ardasener]

The primary goal of this PR is to implement binary readers and writers for the newly added Array format. However the current class structure of the I/O system caused some issues with this. So this PR also features a slight redesign of the system to accommodate Array and make the library more future proof.

Previous Design

template <typename IDType, typename NNZType, typename ValueType>
class BinaryWriter : public Writer<IDType, NNZType, ValueType>,
                     public WritesCOO<IDType, NNZType, ValueType>,
                     public WritesCSR<IDType, NNZType, ValueType> {
public:
  BinaryWriter(std::string filename);
  ~BinaryWriter() = default;
  void WriteCOO(format::COO<IDType, NNZType, ValueType> *coo) const;
  void WriteCSR(format::CSR<IDType, NNZType, ValueType> *csr) const;

private:
  std::string filename_;
};

The main problem with this design is that to write an Array object, the user still has to pass three template types only one of which would be used. Similar situations would occur as new formats with more or less templates get added to the library.

New Design

class Writer {
public:
  virtual ~Writer() = 0;
};


template <typename IDType, typename NNZType, typename ValueType>
class BinaryWriterOrderTwo : public Writer
                     public WritesCOO<IDType, NNZType, ValueType>,
                     public WritesCSR<IDType, NNZType, ValueType> {
public:
  explicit BinaryWriterOrderTwo(std::string filename);
  ~BinaryWriterOrderTwo() = default;
  void WriteCOO(format::COO<IDType, NNZType, ValueType> *coo) const;
  void WriteCSR(format::CSR<IDType, NNZType, ValueType> *csr) const;

private:
  std::string filename_;
};


template <typename T>
class BinaryWriterOrderOne : public Writer
                     public WritesArray<T> {
public:
  explicit BinaryWriterOrderOne(std::string filename);
  ~BinaryWriterOrderOne() = default;
  void WriteArray(format::Array<T> *arr) const;

private:
  std::string filename_;
};

The new design is inspired heavily by the recently updated conversion system. However in this case CRTP is not necessary so it is omitted. For more information, please refer to #85.

[AmroAlJundi]

I think this should be ReadArray, right?

[ardasener]

Yep, totally

[AmroAlJundi]

I think this whole interface is pointless at this point 🤔 what do you think?

[AmroAlJundi]

Maybe we should rename it as ReadOrderTwoFormat since that's what it's supposed to do.

[ardasener]

According to CLion, it is not used anywhere in the library. So I think we should just remove it. Graphs right now have methods for both CSR and COO so a generic one is not really necessary.

[ardasener]

We can always add it back in if we end up needing it.

[AmroAlJundi]

fair enough.

[AmroAlJundi]

Looks good. There's a typo + I think ReadsSparseFormat was meant for reading connectivity in graphs so maybe it should be ReadsOrderTwoFormat. What do you think?