Fix incorrect propagation of field's nullability into its inner list

[teh-cmc]

The arrow pass of the codegen wrongly attributes the nullability of a field to its inner entity (whether it's a list or fixed-size list), which leads to corrupt datatypes being generated.

This generally doesn't cause any issues in practice because one cannot trust what the datatype says anyway, and nullability is instead checked by directly looking at the bitmap itself at runtime.

But, it turns out that the C++ backend completely ignore this piece of metadata at the moment, and therefore ends up generating datatypes that don't match what the Rust & Python backends generate, leading to broken roundtrips.

Fix the semantic pass so everyone agrees once again.

What

Checklist

• I have read and agree to Contributor Guide and the Code of Conduct
• I've included a screenshot or gif (if applicable)
• I have tested demo.rerun.io (if applicable)

• PR Build Summary
• Docs preview
• Examples preview
• Recent benchmark results
• Wasm size tracking

[Wumpf]

great catch, makes sense! extra points for the nice comment