@groupby failed on CategoricalString with missing value

Let's say, we try to @groupby on a CategoricalString column

df = DataFrame(
    fruit    =  ["Apple", "Banana", "Cherry", "Apple", "Banana", "Cherry"],
    price    =  [1.2, 2.0, 0.4, 1.2, 2.0, 0.4])

categorical!(df, :fruit)
df |> @groupby(_.fruit) |> @map({key=key(_), m=mean(_.price)})

The result is exactly as expected.

3×2 DataFrame
│ Row │ key          │ m       │
│     │ Categorical… │ Float64 │
├─────┼──────────────┼─────────┤
│ 1   │ Apple        │ 1.2     │
│ 2   │ Banana       │ 2.0     │
│ 3   │ Cherry       │ 0.4     │

Next, we make the same column contain some missing values.

df = DataFrame(
    fruit    =  ["Apple", missing, "Cherry", "Apple", "Banana", "Cherry"],
    price    =  [1.2, 2.0, 0.4, 1.2, 2.0, 0.4])

categorical!(df, :fruit)
df |> @groupby(_.fruit) |> @map({key=key(_), m=mean(_.price)})

Then, if failed with an error message stating somewhat like this.

ERROR: MethodError: convert(::Type{DataValue{CategoricalString{UInt32}}}, ::CategoricalString{UInt32}) is ambiguous

However, this behavior is weird since String column with similar missing values does not issue any complaints.

df = DataFrame(
    fruit    =  ["Apple", missing, "Cherry", "Apple", "Banana", "Cherry"],
    price    =  [1.2, 2.0, 0.4, 1.2, 2.0, 0.4])

df |> @groupby(_.fruit) |> @map({key=key(_), m=mean(_.price)})

key      │ m
─────────┼────
"Apple"  │ 1.2
#NA      │ 2.0
"Cherry" │ 0.4
"Banana" │ 2.0

This limitation may have to be fixed, since it is natural that String columns are recoded as Categorical columns before beginning any analysis.

Thanks.

Reproduce the two @groupby examples in the issue, comparing a CategoricalString column with missing values to the equivalent String column. Trace the @groupby handling that produces the ambiguous DataValue conversion. Done means the categorical example groups the missing value without the reported MethodError and preserves the expected results.