Replace `remove_na_observed_predicted()` by `na.omit()`

[nikosbosse]

We currently have a function remove_na_observed_predicted() that removes all rows where either predicted or observed is NA.

I think it would be good to simply replace this function by na.omit().

Reasoning:

• Having NA values anywhere in the data (not just in observed and predicted) is potentially risky and could break things (e.g. mess up grouping according to the forecast unit). Handling this makes things more complicated.
• na.omit() is a known function. Having (and exporting) a function that just removes NA values in certain columns seems unnecessarily complicated).

[seabbs]

This sounds fine but does mean that we should be warning users if we are dropping a bunch of their data prior to scoring as otherwise it could have some dangerous silent consequences.

[nikosbosse]

Closing as completed via #465