Enable FP16 thresholding and elementwise ops

[SpiritSeeker]

Enable FP16 support for thresholding layers. The following changes are made

• npy data type for storing inputs and thresholds can now be float16 or float32 instead of default float32
  • changes are specific to thresholding RTL and HLS
• npy2apintstream now supports half datatype (updated cnpy dependency)
  • changes effect all custom ops
• new test with float16 inputs and thresholds

Enable FP16 support for elementwise layers. The following changes are made

• enable storage of inputs and outputs as half
  • changes in HLS backend, elementwise HLS, and data packing
• new test with float16 inputs

[auphelia]

Thank you, @SpiritSeeker !