add layout template parameter to pyarray and pytensor

[wolfv]

This enables fast iterators.

Currently, default template parameter is still dynamic, so no backwards incompatibility.

[SylvainCorlay]

Yeah!

[wolfv]

Note:

I had to add -fvisibility=hidden to the test cmake list to silence some warnings (as described also in the pybind11 FAQ).

Also, there was a warning about the pytensor copy constructor:

/home/wolfv/Programs/xtensor-python/include/xtensor-python/pytensor.hpp: In instantiation of ‘xt::pytensor<T, N, L>::pytensor(const self_type&) [with T = int; long unsigned int N = 3; xt::layout_type L = (xt::layout_type)0; xt::pytensor<T, N, L>::self_type = xt::pytensor<int, 3>]’:
/home/wolfv/Programs/xtensor-python/test/test_pytensor.cpp:102:33:   required from here
/home/wolfv/Programs/xtensor-python/include/xtensor-python/pytensor.hpp:333:12: warning: base class ‘class xt::xcontainer_semantic<xt::pytensor<int, 3> >’ should be explicitly initialized in the copy constructor [-Wextra]
     inline pytensor<T, N, L>::pytensor(const self_type& rhs)
            ^~~~~~~~~~~~~~~~~

which was fixed by swapping the initialization of the base_type to self_type.