complete re-work of code

This release is a complete re-work of underlying code, while the API is mostly unchanged.

Features added:

    support for tensors of rank 4
    broadcasting
    views
    in-place operators
    customisation via a single header file (ulab.h)
    numpy-compatibility for functions, where it makes sense

Functions/methods added

    .tobytes()
    concatenate
    cross
    full
    logspace
    diagonal function