Re-consider IC output for refitted models

Using the latest github version of fable here (but issue is also present on CRAN version).

library(tsibble)
library(fable)
#> Loading required package: fabletools
library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union

lung_deaths <- as_tsibble(mdeaths)

# Fit ARIMA on first part of timeseries.
ts1 <- lung_deaths %>% filter(index <= as.Date("1977-01-01"))
#> Warning in mask$eval_all_filter(dots, env_filter): Incompatible methods
#> ("<=.vctrs_vctr", "<=.Date") for "<="
fit <- ts1 %>%
  model(ARIMA(value ~ 1 + pdq(1,0,0) + PDQ(0,0,0)))

# Refit ARIMA with one more observation
ts2 <- lung_deaths %>% filter(index > as.Date("1977-01-01"))
#> Warning in mask$eval_all_filter(dots, env_filter): Incompatible methods
#> (">.vctrs_vctr", ">.Date") for ">"
fit %>%
  refit(ts2 %>% head(1)) %>%
  report()
#> Series: value 
#> Model: ARIMA(1,0,0) w/ mean 
#> 
#> Coefficients:
#>          ar1  constant
#>       0.7851  364.8801
#> s.e.  0.1030   43.9553
#> 
#> sigma^2 estimated as 1579:  log likelihood=-5.58
#> AIC=13.16   AICc=9.16   BIC=11.16

# Refit ARIMA with two more observations
fit %>%
  refit(ts2 %>% head(2)) %>%
  report()
#> Warning: It looks like you're trying to fully specify your ARIMA model but have not said if a constant should be included.
#> You can include a constant using `ARIMA(y~1)` to the formula or exclude it by adding `ARIMA(y~0)`.
#> Error: Problem with `mutate()` input `ARIMA(value ~ 1 + pdq(1, 0, 0) + PDQ(0, 0, 0))`.
#> x Could not find an appropriate ARIMA model.
#> This is likely because automatic selection does not select models with characteristic roots that may be numerically unstable.
#> For more details, refer to https://otexts.com/fpp3/arima-r.html#plotting-the-characteristic-roots
#> ℹ Input `ARIMA(value ~ 1 + pdq(1, 0, 0) + PDQ(0, 0, 0))` is `(function (object, ...) ...`.

^{Created on 2020-05-18 by the reprex package (v0.3.0)}

Run the supplied R reprex against the CRAN and GitHub versions, focusing on refit() followed by report() for the ARIMA model. No source file or test is named; trace those entry points and how information criteria are calculated for one versus two added observations. Done means the differing output and resulting error have an established, consistent expected behavior.