pandas.Categorical not preserved during DataFrame conversion and jl.convert fails

Hi and thanks for the great package!

While working with `juliacall` + `PythonCall.jl`, I ran into two issues related to pandas.Categorical handling.

(In all examples below, `jl`refers to Main from `juliacall`, i.e., f`rom juliacall import Main as jl`.)

**DataFrame conversion ignores Categorical columns**
When passing a `pandas.DataFrame` with categorical columns (i.e., `dtype='category'`), those columns are silently converted to `Int64` vectors in Julia (presumably the .codes). This results in CategoricalArray semantics being lost — so interactions in Julia formulas like `id & η1` are treated as numeric rather than generating dummy variables.

**`jl.convert()` can’t convert pandas.Categorical to any Julia type**
I tried using `jl.convert(CategoricalArray, col)` directly on a `pandas.Series` with categorical dtype, but got a `MethodError`. It appears `PythonCall` doesn’t yet support converting `pandas.Categorical` to any Julia-native type.

To work around this, I convert the column to `str` in Python (so it arrives as a `Vector{String}`), then manually wrap it in `categorical(...)` on the Julia side. This works, but it's not ideal for type fidelity or automatic translation.

Let me know if there's a cleaner workaround — or if you'd be open to a PR to improve automatic `CategoricalArray` support.

Thanks again!

Start with the pandas.DataFrame conversion path and the jl.convert(CategoricalArray, col) case described in the issue. Trace how pandas.Categorical columns are represented, then verify that conversion preserves CategoricalArray semantics and that direct conversion no longer fails.

Do mô hình lập chỉ mục viết ra từ nội dung của issue.