Windows issue with negative decimals

Not sure if this is a nanoarrow issue or with the PyArrow export, but figured I'd start here.

When consuming an array like this:

```python
schema = pa.schema([("col", pa.decimal128(38, 10))])
tbl = pa.Table.from_arrays([
pa.array(["1234567890.123456789", "-99876543210.987654321",])
], schema=schema)
```

and using `ArrowDecimalAppendDigitsToBuffer` in nanoarrow, it seems like negative values on windows only are not properly represented.

I created an "MRE" for this in this library:

https://github.com/WillAyd/capsule_testing/tree/739bc608b96101ada424ef6d9736e2dfffcc2f11

You can see the issue for Windows in CI logs:

https://github.com/WillAyd/capsule_testing/actions/runs/10482665140/job/29034136906#step:4:358

Start with ArrowDecimalAppendDigitsToBuffer and the linked capsule_testing MRE, then compare its Windows CI output with the expected negative decimal values. Reproduce the issue on Windows and trace the decimal formatting path; done means negative decimals are represented correctly there without changing the existing positive-value behavior.

Written by the indexing model from the issue text.