Optionally stop data deletion after compute

In the following locs I'm seeing that the module is deleting the data with `del self.data` can we have a flag that avoids this deletion so that we can drill down into good/bad data points at a later stage? I can raise a raise a PR if you think this is a good idea.

https://github.com/huggingface/evaluate/blob/55f1bc6e072b05c2d9db1589a07e20f38902b1ec/src/evaluate/module.py#L471

```python
...
if self.buf_writer is not None:
self.buf_writer = None
del self.data
self.data = None
else:
# Release locks and delete all the cache files. Process 0 is released last.
for filelock, file_path in reversed(list(zip(self.filelocks, self.file_paths))):
logger.info(f"Removing {file_path}")
del self.data
self.data = None
del self.writer
self.writer = None
os.remove(file_path)
filelock.release()
...
```