Join errors when retry token is unavailable

[NetcraftPm]

If we fail to acquire a retry token, we are currently discarding the original error. According to the Retry Behaviour reference guide, we are expected to return the actual response (error) in this situation.

With these changes, we now return the joined error, which allows callers to inspect and match on the error that caused the request to fail.

[lucix-aws]

This seems ok but shouldn't the retry error wrap the original error instead?

[NetcraftPm]

This seems ok but shouldn't the retry error wrap the original error instead?

I'm happy to make it wrap the error in GetRetryToken instead if you prefer. My logic was that it seemed that given the method name is GetRetryToken it doesn't make much sense for it to wrap the error it's given, as that's conceptually to determine what (size of) token to reserve. We could also use fmt.Errorf to join the errors, but it feels like that would make the actual error string quite hard to read.