Commands in the `release` module always return exit code 0, even if they fail

Hi,
When executing commands such as `nbdev_pypi` from a shell (which get routed to `nbdev.release:release_pypi`), the commands always return exit code 0 regardless of the errors happening inside the command.

I ran into this issue while using `nbdev_pypi` inside my CI/CD. If I forget to bump the version, for example, the `twine upload` command will fail with a "409 Conflict" error, but the `nbdev_pypi` command still returns error code 0 (success).
This is problematic, because there is no way to catch this error and do something useful (like fail the CI/CD build).

The code for `release_pypi` uses os.system calls like:
```python
system(f'twine upload --repository {repository} {_dir}/dist/*')
```
https://github.com/fastai/nbdev/blob/4af4d479c78880f4a18af4254b119f8af8b3a8a4/nbdev/release.py#L307-L313C68

As far as I can see, the call to `os.system()` returns the exit code as a 0/1 but does not throw an exception.

I'm happy to contribute a pull request to fix this, but would like some guidance on what is the right way to go about it.
Options I see:
1. Change code to use subprocess instead of os.system()
2. Keep using os.system, but instead of calling it directly - add a wrapping function like execute_throw_on_error() which will check the response code and throw exception if it's not 0.
3. Somehow handle it in a more generic way in the the call_parse decorator (but what do we do if there are multiple calls to os.system and we want to fail if either fails?)

Thanks!

Julian.

Start in nbdev/release.py around release_pypi and its os.system call for `twine upload`. Inspect the other release commands for the same exit-status handling, then reproduce a failed upload such as the reported 409 Conflict. Done means a failing command propagates a nonzero exit code so CI/CD can detect the failure.

Written by the indexing model from the issue text.