Rework quantile scores

[nikosbosse]

Update: I reworked the part about bias_quantile() again in #396. Maybe it would be easiest to just ignore quantile_bias() for now? I can also merge #396 into this one (although there are some updates in between) or I could replicate the changes in #396 here to make reviewing it a bit easier.

This PR

• creates a version of the WIS that
  • accepts a quantile input
  • returns one single score for a forecast (comprising multiple predictive quantiles, see image below)
• Creates a function that computes the WIS for a one-to-one relationship
  • Meaning that it multiple scores for one forecast, namely one score per quantile
  • This is the behaviour of score() at the moment where you get one score per quantile. After talking to Seb it's a bit unclear whether we want / need that.
  • So at the moment the function just sits around. I created an issue to decide what to do with it. Also the different output formats seem a bit crazy, so maybe there is no need to review it now until we made a decision how it fits in (can also remove it from this PR if you prefer).
  • Issue related to that is Decide how WIS should be computed and returned #386
• Adds input checks for forecasts in an interval format (which we need to think about whether or not to support an in what way, again related to Decide how WIS should be computed and returned #386)
• Update bias_quantile() to work with several observations instead of only one (accepting the same input formats as wis(). This makes use of a function bias_quantile_single() which ultimately could be replaced by just data.table code in a single function. In the spirit of not cramming to many things in here (which I appreciate I completely failed to do yet again) I created an issue for that Merge bias_quantile() and bias_quantile_single() #387
• Update examples and tests

Next step in a future PR is then to update score.quantile() to work with these functions.

[nikosbosse]

@jhellewell14 maybe this is not the most terrible PR to have a look at if you feel inclined

[seabbs]

Do you want this reviewed @nikosbosse?

[nikosbosse]

Yes please!

[seabbs]

If you could clean this up into the actual proposal that would be good as trying to review across two PRs without sufficient information is very hard.

[seabbs]

I think this is effectively unreviewable in its current form and given where we are with the rest of the development we should merge and aim to improve the use of planning issues and modular PRs going forward.