Validate f function in epi[x]_slide; give better feedback if doesn't take enough args

[brookslogan]

See this example:

epi_slide(jhu_csse_county_level_subset, function(x) print(x))
Error in .f(.x, ...) : unused argument (list())

The error message isn't that informative, and seems like it might be fairly commonly encountered. (I have trouble remembering the g arg despite discussing it for a while.)

We can check that f takes enough arguments beforehand. I believe I have something along these lines for a different function in #102. (See fn_max_n_args definition and use.)

[brookslogan]

I've hit this while trying to use lm.fit or slice_min as the computation, which may give more confusing error messages than the above.

[brookslogan]

Having checks also hit the above cases might require something more detailed than checking fn_max_n_args, such as requiring the number of positional args before seeing any ... to be at least the number of mandatory arguments for the slide. (This might require explicitly forwarding some things to comp_one_group that previously we forwarded via ..., but might catch slice_min even though it takes more than enough args.)

[nmdefries]

I've hit this while trying to use lm.fit or slice_min as the computation

Both lm.fit and slice_min take two positional args before ..., and are successfully passed x and g by epi_slide. The errors

epi_slide(jhu_csse_county_level_subset, slice_min, before=2L)
# Error in `.f()`:
# ! Problem while computing indices.
# Caused by error in `UseMethod()`:
# ! no applicable method for 'group_keys' applied to an object of class "rlang_data_pronoun"

epi_slide(jhu_csse_county_level_subset, lm.fit, before=2L)
# Error in .f(.x, ...) : incompatible dimensions

both come from internal arg validation by the provided function, so this behavior seems correct. (Also, since both functions already take 2 args before ... the approach proposed above wouldn't catch them; we could check arg names?)

[nmdefries]

Closed in #302