add cossin LUT

[matthuszagh]

Implements a fast cossin computation using an 8-bit LUT. This achieves better than 0.9LSB maximum error for 16-bit output. It uses a 20-bit integer phase input and produces 18-bit integer outputs.

Considerations

• There appears to be a small phase offset/bias since the error produces an oscillatory pattern. However, the performance/accuracy appear to be good enough so this shouldn't present a problem. A plot of the error values as a function of phase is shown below.

• It may make sense to make shift_round integer-generic. I believe the easiest way to do this would be to add a dependency on num.

Related

• sin/cos/atan2/sqrt implementation analysis #191