Verify request retry behaviour of requests about stream-python

[tbarbugli]

few thoughts:

• timeouts should be quick for production environments
• timeouts should not be too aggressive on development
• we should not retry failed responses (eg. 4xx, 5xx)
• aggressive timeouts + retry might be a good fit for lot of production applications

[matthisk]

For fine grain control over retries we should use urllibr Retry class. When passing an instance of this class to max_retries on HTTPAdapter you can control retry behavior.

According to the docs this behavior only applies to failed connections and timeouts, but I guess if you supply an instance of the Retry class you can have more control:

max_retries (int) – The maximum number of retries each connection should attempt. Note, this applies only to failed connections and timeouts, never to requests where the server returns a response.