stat_smooth() drops failed groups

[teunbrand]

This PR aims to fix #5352.

Briefly, it wraps the fitting/predicting procedure in a tryCatch() block to rethrow errors as warnings and drops the group for which the procedure fails. This prevents one bad group stopping the fitting in all other groups.

The example from #5352 with this PR (in particular, see the very last warning):

devtools::load_all("~/packages/ggplot2")
#> ℹ Loading ggplot2
#> Warning: package 'testthat' was built under R version 4.3.1

diamonds |> 
  subset(cut == "Ideal" & color == "J") |>
  ggplot( aes(carat, price)) +
  geom_point(aes(color = clarity)) +
  geom_smooth(aes(color = clarity))
#> `geom_smooth()` using method = 'loess' and formula = 'y ~ x'
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric = parametric,
#> : span too small.  fewer data values than degrees of freedom.
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric = parametric,
#> : at 0.9598
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric = parametric,
#> : radius 0.00010404
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric = parametric,
#> : all data on boundary of neighborhood. make span bigger
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric = parametric,
#> : pseudoinverse used at 0.9598
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric = parametric,
#> : neighborhood radius 0.0102
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric = parametric,
#> : reciprocal condition number 1
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric = parametric,
#> : at 3.0202
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric = parametric,
#> : radius 0.00010404
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric = parametric,
#> : all data on boundary of neighborhood. make span bigger
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric = parametric,
#> : There are other near singularities as well. 0.00010404
#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric = parametric,
#> : zero-width neighborhood. make span bigger

#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric = parametric,
#> : zero-width neighborhood. make span bigger
#> Warning: Failed to fit group 1.
#> Caused by error in `predLoess()`:
#> ! NA/NaN/Inf in foreign function call (arg 5)

^{Created on 2023-07-27 with reprex v2.0.2}

[teunbrand]

As an aside, is there some smart way to reduce repeated parts in the warnings? In particular, the call getting mentioned several times is a bit annoying and it would seem cleaner to just warn like so:

#> Warning in simpleLoess(y, x, w, span, degree = degree, parametric = parametric,
#> : span too small.  fewer data values than degrees of freedom.
#> : at 0.9598
#> : radius 0.00010404
#> : all data on boundary of neighborhood. make span bigger
# etc...
#> Warning: Failed to fit group 1.
#> Caused by error in `predLoess()`:
#> ! NA/NaN/Inf in foreign function call (arg 5)

[thomasp85]

I'm not sure what we can do about the excessive warnings. Summarising them would require us to make some very hard expectations about the warnings coming from the call...

[thomasp85]

LGTM