Py torch2 bugfix - Githubissues

I fixed some bugs appearing with PyTorch 2.x. The whole setup is backwards compatible with pytorch 1.11 and 1.12 - the tests run and pass.

I had to edited some older tests that fail due to precision and inserted an almost_equal where appropriate.
- One test, test_FermionMatrix_hermitian, is still failing. I don't understand why. When printing the difference
```
NSL::LinAlg::max( NSL::LinAlg::abs(MdaggerM - MdaggerMH) )
```
  the largest error appears to be 1e-16 on multiple runs. But the checkup
```
REQUIRE(almost_equal(MdaggerM, MdaggerMH, std::numeric_limits<Type>::digits10-2).all());
```
  still fails for various instances.
I added tests for NSL:Linalg::inner_product which was failing with the new PyTorch version.
Adding chemical potential in NSL::FermionMatrix::HubbardExp caused some bugs with the call order (beta vs mu) so I refactored this putting mu as the last argument of the constructor.

Marcel-Rodekamp / NSL