Cast large_string to string right before saving to Parquet.

[davidknight-seequent]

Description

Second attempt at getting rid of large_string.

• Reverted previous changes meant to fix this problem.
• Targeted functions that actually save dataframes to Parquet.

Checklist

• I have read the contributing guide and the code of conduct

[wordsworthc]

My understanding here is that the issue lies in the default handling of pandas StringDtype when converting a pandas dataframe to a pyarrow table. To avoid this area of code becoming a patchwork quilt, I think we should

1. revert the original change that didn't fix the issue; then
2. update the way we convert dataframes to tables
   a. df.convert_dtypes(dtype_backend="pyarrow") seems to be one way of handling this, but check for performance penalties first.

Optionally, add some kind of documentation about this issue with pandas v3, because this seems like an easy trap to fall into if you're doing DIY conversion

[davidknight-seequent]

The convert_dtypes approach is about 2.3x slower than the approach implemented in this PR. This is because it copies and re-infers every column's dtype to pandas before handing off to Arrow.