Consider not trying to uninstall helm releases that weren't successfully installed

[craigwalton-dsit]

E.g. inspect asks us to install 100 releases, we queue them up and before they're finished, user ctrl+c's and inspect asks us to uninstall them all, then we show a bunch of errors. Can we check if the helm releases are in our "tracked" list before attempting uninstall?

[craigwalton-dsit]

Currently, we have

    async def install(self, release: Release) -> None:
        """
        Installs a release and tracks it for eventual cleanup.

        Args:
          release (Release): The release to install and track.
        """
        # Track the release regardless of the install result.
        self._installed_releases.append(release)
        await release.install()

We add the release to self._installed_releases() before we've actually successfully installed it because:

1. The install might fail e.g. Helm timeout in which case, an exception will be raised, but the release won't be uninstalled by Helm (unless we passed --atomic flag to helm install)
2. If the install gets cancelled by Inspect (e.g. user hits ctrl+c or maybe even if there's an eval failure) then an asyncio.CancelledError is raised. We don't really know at what stage the cancellation was at though - was it whilst waiting for the semaphore, or did we actually get as far as starting helm install?

In case 1, it is correct to add it to self._installed_releases.
In case 2, if the cancellation happened before calling helm install we could avoid adding it to the list. But determining that is not trivial.