Restrict dtypes, bugfixes, test basic tensor functions

[tostenzel]

• bugfixes

• due to dtype related bugs:

  • move dtypes to own module
  • only allow bool, int32 and float32
    - Only allow float32 when Tensor(list). (Perhaps allow bool, int32, too if list is only bool, and only int32 in the future?) -> Yet most operations in TensorData cast to float32 anyways.

• ingetration test for mnist test acc > 93 %

• unit tests for tensor distributed over according modules

[tostenzel]

TODO:

• Perhaps split test_tensor.py across multiple modules according to repo structure
• Figure out how to adjust tests to restricted dtypes