modify automatic resolution detection for irregular data?

[fabian-s]

should constructors use curve-specific arg-values instead of merged ones for figuring out appropriate resolution?

• + avoid super fine resolution for data with minimally "jiggled" argvalues
• - downstream complications:
  • less precise representation of irregular data (e.g. : this could mess up effects of small / subtle warpings....)
  • likely to fail for very sparse/irregular data (will at least need additional checks to find a good compromise with what "merged" grid would yield as resolution)

simpler idea: just inform users if irregular input yields "too fine" resolution and let them set resolution manually.

related: tidyfun/tf#9

[fabian-s]

see utils.R: get_resolution() -- we use "smallest observed interval between consecutive evaluations"/10, rounded down to the nearest power of 10, seems like a reasonable default.