Gracefully handle nans in the input data

[djinnome]

This might be a problem for @fivegrant but if we receive nans in the dataset that is given to us, we shouldn't try to interpret it, we should instead drop those calibration events.

[djinnome]

We already handle missing data, we should handle nans in the same way.

[SamWitty]

Closing, as this is out-of-scope for our service. We assume that NANs will have been caught and pruned in the TA4 service. We can however return an error if we see NANs.