type_convert() does not parse IEEE 754 double values (NaN, Inf, -Inf)

[khusmann]

Loading a csv with NaN, Inf, -Inf values with col_double() works as expected (#1225):

read_csv("a\nNaN\nInf\n-INF", col_types = cols(a=col_double()))
## A tibble: 3 × 1                                                                                                                                                         
#      a
#  <dbl>
#1   NaN
#2   Inf
#3  -Inf

But a similar type_convert() throws errors, and loads all the special values as NA:

type_convert(tibble(a=c("NaN", "Inf", "-INF")), col_types = cols(a=col_double()))
## A tibble: 3 × 1
#      a
#  <dbl>
#1    NA
#2    NA
#3    NA
#Warning messages:
#1: [0, 1]: expected a double, but got 'NaN' 
#2: [1, 1]: expected a double, but got 'Inf' 
#3: [2, 1]: expected a double, but got '-INF'

Is this the intended behavior? Or an artifact of v1 vs v2 (#1277)? Thanks!