Add support for bool, double, int32, uint32 and float32 on Tensors via TensorT

[axsaucedo]

Fixes #2
Fixes #144

This PR explores the approaches that could be introduced to support multiple types for tensors - it includes:

• Conversion of top level Tensor class as just a high level container that doesn't store memory
• TensorT class implementation that introduces abstraction to provide data in CPU with std::vector
• All other classes can use the top level Tensor class without any modifications
• Updated array_classification and logistic regression example showing how it works at the moment
• Introduces kp::Tensor::TensorDataTypes which provide explicit data information

Further points to explore:

• Clean up code and move TensorT to its own hpp / cpp files
• Potentially drop completely the mData parameter as discussed in Explore removing std::vector mData; completely from Tensor in favour of always storing data in hostVisible buffer memory (TBC) #144 in favour of memory copy on read
  • This would require exploration of potentially using non-owning container like span to avoid passing around pointer memory size
• Python bindings to align with class specific tensors, ideally using numpy dtypes
• Add documentation

Other things to explore in separate issue / pr:

• Exploration of validation for shaders (most probably outside of scope of PR)

[alexander-g]

Looks good.
I am already using different data types in my project via .view('uint32'). This was quite a pain to get it working correctly, so if Kompute supports this from the beginning it would of course be nice.