Add an exponential backoff to the retry function

This PR adds a 200% exponential backoff to the default retry function. This backoff only applies in situations where the Rate Limit header doesn't exist, which should be vanishingly few situations, but unfortunately does exist within the SaaS environment of GitLab: gitlab-org/gitlab#365728

Without the backoff, all 5 default retries would be exhausted in < 1-2 seconds, which would likely result in 429 errors being passed upstream since the API limits on SaaS are measured in requests per minute. With the backoff, the 100ms backoff translates to 3.2 seconds on the final wait (which isn't horrible). If a user increased the retry up to 10 requests, the final retry would wait just shy of 2 minutes.

This feels like a reasonable low impact change that would result in a good quality of life improvement for most users of the library :)

I also debated making the exponent configurable, but I'm not sure that's necessary, since you can already control the min and the max.