laion2B-en failed to load on Windows with PrefetchVirtualMemory failed

### Describe the bug

I have downloaded laion2B-en, and I'm receiving the following error trying to load it:
```
Resolving data files: 100%|██████████| 128/128 [00:00<00:00, 1173.79it/s]
Traceback (most recent call last):
File "D:\Art-Workspace\src\artworkspace\tokeneval\compute_frequencies.py", line 31, in
count = compute_frequencies()
^^^^^^^^^^^^^^^^^^^^^
File "D:\Art-Workspace\src\artworkspace\tokeneval\compute_frequencies.py", line 17, in compute_frequencies
laion2b_dataset = load_dataset("laion/laion2B-en", split="train", cache_dir=_CACHE_DIR, keep_in_memory=False)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\bberman\Documents\Art-Workspace\venv\Lib\site-packages\datasets\load.py", line 2165, in load_dataset
ds = builder_instance.as_dataset(split=split, verification_mode=verification_mode, in_memory=keep_in_memory)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\bberman\Documents\Art-Workspace\venv\Lib\site-packages\datasets\builder.py", line 1187, in as_dataset
datasets = map_nested(
^^^^^^^^^^^
File "C:\Users\bberman\Documents\Art-Workspace\venv\Lib\site-packages\datasets\utils\py_utils.py", line 456, in map_nested
return function(data_struct)
^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\bberman\Documents\Art-Workspace\venv\Lib\site-packages\datasets\builder.py", line 1217, in _build_single_dataset
ds = self._as_dataset(
^^^^^^^^^^^^^^^^^
File "C:\Users\bberman\Documents\Art-Workspace\venv\Lib\site-packages\datasets\builder.py", line 1291, in _as_dataset
dataset_kwargs = ArrowReader(cache_dir, self.info).read(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\bberman\Documents\Art-Workspace\venv\Lib\site-packages\datasets\arrow_reader.py", line 244, in read
return self.read_files(files=files, original_instructions=instructions, in_memory=in_memory)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\bberman\Documents\Art-Workspace\venv\Lib\site-packages\datasets\arrow_reader.py", line 265, in read_files
pa_table = self._read_files(files, in_memory=in_memory)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\bberman\Documents\Art-Workspace\venv\Lib\site-packages\datasets\arrow_reader.py", line 200, in _read_files
pa_table: Table = self._get_table_from_filename(f_dict, in_memory=in_memory)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\bberman\Documents\Art-Workspace\venv\Lib\site-packages\datasets\arrow_reader.py", line 336, in _get_table_from_filename
table = ArrowReader.read_table(filename, in_memory=in_memory)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\bberman\Documents\Art-Workspace\venv\Lib\site-packages\datasets\arrow_reader.py", line 357, in read_table
return table_cls.from_file(filename)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\bberman\Documents\Art-Workspace\venv\Lib\site-packages\datasets\table.py", line 1059, in from_file
table = _memory_mapped_arrow_table_from_file(filename)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\bberman\Documents\Art-Workspace\venv\Lib\site-packages\datasets\table.py", line 66, in _memory_mapped_arrow_table_from_file
pa_table = opened_stream.read_all()
^^^^^^^^^^^^^^^^^^^^^^^^
File "pyarrow\ipc.pxi", line 757, in pyarrow.lib.RecordBatchReader.read_all
File "pyarrow\error.pxi", line 91, in pyarrow.lib.check_status
OSError: [WinError 8] PrefetchVirtualMemory failed. Detail: [Windows error 8] Not enough memory resources are available to process this command.
```

This error is probably a red herring: https://stackoverflow.com/questions/50263929/numpy-memmap-returns-not-enough-memory-while-there-are-plenty-available In other words, the issue is related to asking for a memory mapping of length N > M the length of the file on Windows. This gracefully succeeds on Linux.

I have 1024 arrow files in my cache instead of 128 like in the repository for it. Probably related. I don't know why `datasets` reorganized/rewrote the dataset in my cache to be 1024 slices instead of the original 128.

### Steps to reproduce the bug

```
# as a huggingface developer, you may already have laion2B-en somewhere
_CACHE_DIR = "."

from datasets import load_dataset
load_dataset("laion/laion2B-en", split="train", cache_dir=_CACHE_DIR, keep_in_memory=False)
```

### Expected behavior

This should correctly load as a memory mapped Arrow dataset.

### Environment info

- `datasets` version: 2.15.0
- Platform: Windows-10-10.0.20348-SP0 (this is windows 2022)
- Python version: 3.11.4
- `huggingface_hub` version: 0.19.4
- PyArrow version: 14.0.1
- Pandas version: 2.1.2
- `fsspec` version: 2023.10.0