Bug: ggplotly ignores colours in ggplot2::scale_fill_gradientn when limits are below the range of the data

This is an odd bug (reported first as a heatmaply issue).

If we create a ggplot2 object with colors based on scale_fill_gradientn, then ggplotly respect the colors. But if the limits are set to be within (instead of outside) the range of the data, then ggplotly stops respecting the colors. This may be some fallback for how to deal with NA colors (but I don't know enough about ggplotly to know for sure).

Here is a self contained example:

library(ggplot2)
library(viridis)
library(plotly)
# Equivalent fill scales do the same job for the fill aesthetic
p <- ggplot(faithfuld, aes(waiting, eruptions)) +
  geom_raster(aes(fill = density)) +
  scale_fill_gradientn(colours = viridis(10), limits = c(0.001,.01))
p
ggplotly(p)

Plot in ggplot2:

Plot in plotly (see the wrong colors produced!)

Start by running the self-contained R example with ggplot2, viridis, and plotly, then compare the ggplot2 and ggplotly color output when scale_fill_gradientn limits are below the data range. Done means ggplotly preserves the intended gradient colors for this case.