replace axis=x|y with kind=percentile|threshold

[wholmgren]

I find it to be nearly impossible to talk about ProbabilisticForecasts with axis='x'|'y'. You really need experience with the arbiter code and probabilistic forecasts to easily reason about what you're reading, too. It's much easier for me to say fixed percentiles or fixed thresholds, and I suspect it would be easier to reason about. So my proposal is to replace the axis='x'|'y' argument with kind='percentile'|'threshold'.

[dplarson]

I support this idea, including the nomenclature of "percentile" and "threshold".

[wholmgren]

Also relevant from Bouallegue et al reference in #520:

Figure 1 shows an example of a cumulative distribution function F(x). A threshold ω and the associated probability forecast 1 − pω as well as a probability level τ and the associated quantile forecast qτ are shown on the plot.

Suggests that perhaps constant_values should be fixed_values.