None values are skipped when reading jsonl in subobjects

### Describe the bug

I have been fighting against my machine since this morning only to find out this is some kind of a bug.

When loading a dataset composed of `metadata.jsonl`, if you have nullable values (Optional[str]), they can be ignored by the parser, shifting things around.

E.g., let's take this example

Here are two version of a same dataset:

[not-buggy.tar.gz](https://github.com/user-attachments/files/16333532/not-buggy.tar.gz)
[buggy.tar.gz](https://github.com/user-attachments/files/16333553/buggy.tar.gz)

### Steps to reproduce the bug

1. Load the `buggy.tar.gz` dataset
2. Print baseline of `dts = load_dataset("./data")["train"][0]["baselines]`
3. Load the `not-buggy.tar.gz` dataset
4. Print baseline of `dts = load_dataset("./data")["train"][0]["baselines]`

### Expected behavior

Both should have 4 baseline entries:
1. Buggy should have None followed by three lists
2. Non-Buggy should have four lists, and the first one should be an empty list.

One does not work, 2 works. Despite accepting None in another position than the first one.

### Environment info

- `datasets` version: 2.19.1
- Platform: Linux-6.5.0-44-generic-x86_64-with-glibc2.35
- Python version: 3.10.12
- `huggingface_hub` version: 0.23.0
- PyArrow version: 16.1.0
- Pandas version: 2.2.2
- `fsspec` version: 2024.3.1