Compression kwargs have no effect when saving datasets as csv

### Describe the bug

Attempting to save a dataset as a compressed csv file, the compression kwargs provided to `.to_csv()` that get piped to panda's `pandas.DataFrame.to_csv` do not have any effect - resulting in the dataset not getting compressed.

A warning is raised if explicitly providing a `compression` kwarg, but no warnings are raised if relying on the defaults. This can lead to datasets secretly not getting compressed for users expecting the behaviour to match panda's `.to_csv()`, where the compression format is automatically inferred from the destination path suffix.

### Steps to reproduce the bug

```python
# dataset is not compressed (but at least a warning is emitted)

import datasets

dataset = datasets.load_dataset("rotten_tomatoes", split="train")

dataset.to_csv("uncompressed.csv")
print(os.path.getsize("uncompressed.csv")) # 1008607

dataset.to_csv("compressed.csv.gz", compression={'method': 'gzip', 'compresslevel': 1, 'mtime': 1})
print(os.path.getsize("compressed.csv.gz")) # 1008607
```

```shell
>>>
RuntimeWarning: compression has no effect when passing a non-binary object as input.
csv_str = batch.to_pandas().to_csv(
```

```python
# dataset is not compressed and no warnings are emitted

dataset.to_csv("compressed.csv.gz")
print(os.path.getsize("compressed.csv.gz")) # 1008607

# compare with
dataset.to_pandas().to_csv("pandas.csv.gz")
print(os.path.getsize("pandas.csv.gz")) # 418561
```

---

I think that this is because behind the scenes `pandas.DataFrame.to_csv` is always called with a buf-like `path_or_buf`, but users that are providing a path-like to `datasets.Dataset.to_csv` are likely not to expect / know that - leading to a mismatch in their understanding of the expected behaviour of the `compression` kwarg.

### Expected behavior

The dataset to be saved as a compressed csv file when providing a `compression` kwarg, or when relying on the default `compression='infer'`

### Environment info

`datasets == 2.13.1`