pie chart color mapping doesn't work properly when aggregating data

This seems to be maybe a partial regression related to #1375. Using similar code to that issue:

library(plotly)
plot_ly(mtcars[c(1,3,5), c("cyl", "drat")], labels = ~cyl, values = ~drat, type = 'pie', 
        marker = list(colors = c("cyan", "magenta", "black"))) %>%
    layout(xaxis = list(showgrid = FALSE, zeroline = FALSE, showticklabels = FALSE),
           yaxis = list(showgrid = FALSE, zeroline = FALSE, showticklabels = FALSE))

plot_ly(mtcars[, c("cyl", "drat")], labels = ~cyl, values = ~drat, type = 'pie', 
        marker = list(colors = c("cyan", "magenta", "black"))) %>%
    layout(xaxis = list(showgrid = FALSE, zeroline = FALSE, showticklabels = FALSE),
           yaxis = list(showgrid = FALSE, zeroline = FALSE, showticklabels = FALSE))

It seems to apply the default palette color to the first group.

Interestingly, if you bang it in there a few more times it shows (though with different order).

plot_ly(mtcars[, c("cyl", "drat")], labels = ~cyl, values = ~drat, type = 'pie', 
        marker = list(colors = c("cyan", "magenta", "black", "magenta", "magenta"))) %>%
    layout(xaxis = list(showgrid = FALSE, zeroline = FALSE, showticklabels = FALSE),
           yaxis = list(showgrid = FALSE, zeroline = FALSE, showticklabels = FALSE))

Start by running the two provided plot_ly examples with mtcars and compare the pie slice colors for aggregated and unaggregated data. Review issue #1375 and trace the color handling for pie charts; done when the supplied colors consistently map to the correct groups without requiring extra color entries.