write_csv() uses overly large precision with some floating-point numbers

[krlmlr]

This seems rare but surprising. Tested with 70d0eb4.

library(readr)
write_csv(data.frame(a = 1.3498), stdout())
#> a
#> 1.3498000000000001
write_csv(data.frame(a = 2.6996), stdout())
#> a
#> 2.6996000000000002
identical(1.3498000000000001, 1.3498)
#> [1] TRUE
identical(2.6996000000000002, 2.6996)
#> [1] TRUE

^{Created on 2022-10-01 with reprex v2.0.2}

[MarekGierlinski]

I think I stumbled upon the same issue in a different context. See #1502.