Consider having range mark derive min/max from data when not given

[mwaskom]

Marks that are based on min/max (i.e. Range or Band) assume that {var}min/{var}max variables are present. (Currently they fail with a pretty unhelpful error if not present, which we should generally fix). If those variables are not present, should these marks represent the entire range of the value variable?

This would be useful in contexts like the planned Percentiles stat; we want that to be able to reduce a distribution to an arbitrary number of percentiles (to show with a dash mark or similar), but we also want to be able to show percentile ranges. Clumsy idea like Percentiles([25, 75], range=True) are possible, the proposed behavior would simplify that. I'm not sure if there are any potential downsides to it.