Evaluate effects of FP16 visibilities on imaging & calibration in support of the 'half-precision' ECP

This work will investigate accuracy or performance effects of utilising FP-16 visibilities in place of FP-32 visibilities in the SDP.

Specifically it will modify the NZAPP prototype imaging/calibration pipeline to accept FP-16 visibilities (but accumulated on an FP-32 grid) and predict FP-16 visibilities.

<TODO> Discuss: Need a better idea of what can be done in PI8

• If we are going to do this it should characterise the worse case effects for FP16