Cleanup redis cluster error handling

[rukai]

When writing our error handling we've settled on the approach of:

• use anyhow for everything, making strong use of its .context() functionality to report chains of errors
• when we need to report errors in a way that allows the caller to decide what to do we use a custom enum instead, possibly containing an Other(anyhow::Error) variant.

However this redis code was written a long time ago and doesnt follow that, so I am updating it to do so in this PR.

TransformError::IO etc are not used for any logic so I have instead combined them into TransformError::Other

Same with ConnectionError I created a transparent ConnectionError::Other variant for them

There was a duplicate tracing::warn in many cases so I removed it and let the caller deal with warning when it needs to.