Add FP16 capablity

[marty1885]

WIP:

This PR aims to:

• [x] Add support to create FP16 tensor
• [x] Add FP16 tensor operators
• [x] Optimize HTM algorithm against FP16
• [x] Clean up how tensor's properties are checked in the backend
  • Reduce LoC, more readable

To both CPU and OpenCL backend.

And for OpenCL exclusively:

• [x] Check the given OpenCL device can process FP16

Tests for the type system

• [x] Check types have the correct size
• [x] Check resulting type of unitary operations
• [x] Check resulting type of binary operations
  • Besides comparison ops, I;m lazy
• [x] Check resulting type of general tensor operations
• [x] Somehow disable the fp16 tests when GPU doesn't support FP16
• [x] A way to convert SP/TM from float to float16
  • Just cast the permeance!

Serialize:

• [x] Save/load fp16 tensors

Ref #12

[marty1885]

Done. Merging!