`colors` argument to `add_trace` is only respected if the trace type *isn't* specified in the `plot_ly` call

I'd expect these two lines of code to produce the same output, but only the first does what I expect (uses the colors argument to pick a palette for the points):

plot_ly(data = iris) %>%
  add_trace(x = ~Sepal.Length, y = ~Sepal.Width, color = ~Species, colors = "Blues", mode = "markers")

plot_ly(data = iris, type="scatter") %>%
  add_trace(x = ~Sepal.Length, y = ~Sepal.Width, color = ~Species, colors = "Blues", mode = "markers")

This seems backwards since I added the type="scatter" to avoid the warning about guessing the trace type, only to have my palette disappear.

Curiously, this behavior isn't present if everything is specified in the same call:

plot_ly(data = iris, x = ~Sepal.Length, y = ~Sepal.Width, color = ~Species, colors = "Blues")
plot_ly(data = iris, x = ~Sepal.Length, y = ~Sepal.Width, color = ~Species, colors = "Blues", type="scatter")

both return

Is colors the correct way to pass a palette to a specific layer? In other places it seems like it's actually colorspace? I know scatter doesn't have a colors argument in schema() (e.g. #1699) but clearly it's working when the type is inferred and I would like to provide a palette...

Reproduce the two plot_ly/add_trace examples with iris and compare the generated palettes. Trace color handling through the plot_ly and add_trace entry points, especially the explicit type="scatter" path. Done means both call forms honor colors="Blues" consistently, with regression coverage for the differing examples.