The main two goals of this PR are:

• don't retry requests to Druid when it indicates it's not going to work;
• forward the error details provided by Druid to the library user.

To achieve the first goal, we have to override go-retryablehttp's CheckRetry function, adding cases when the request should not be retried. This depends not only on the response status code from Druid, but also on the response body, as indicated from the different error codes detailed in Druid's documentation.

To handle the second goal, both CheckRetry and ErrorHandler need to be overriden. CheckRetry needs to return the error received from Druid, and ErrorHandler needs to pass this error forward, as well as perform some clean-up.