How to bypass automatic conversion to Float64?

I want to get my hands on a `Tuple{Float32, Float32}` in Python and there doesn't seem to be a good way of doing it

```pycon
>>> jl("typeof")(jl("Tuple{Float32, Float32}")((0.0, 1.0)))
Julia: Tuple{Float64, Float64}
```

Reproduce the PythonCall expression shown in the issue and inspect the conversion path used when constructing Tuple{Float32, Float32} from Python. The work is done when a Python-side tuple can preserve Float32 elements and the resulting Julia type matches the requested tuple type.

Written by the indexing model from the issue text.