@map followed by @replacena with a function defined using a macro

It seems that column transformation functions defined using macros can behave differently than functions defined without macros. Here's an example:

using Match, DataFrames, Query
df = DataFrame(a = [1, 2], b = [3, missing])
f(x) = @match x begin 
    3 => "three"
    _ => missing
end
df |> 
    @map({_..., c = f(_.b)}) |>
    @replacena("successfully replaced")

The output still has NAs in it, despite the final @replacena command.

If I replace f with an equivalent version without macros (even just the one obtained by expanding the macro), this works as expected.

Start by running the issue's Julia example with Match, DataFrames, and Query, focusing on the pipeline combining @map and @replacena. Compare the macro-defined function with its expanded equivalent and verify that the final output replaces all missing values.