@retryable should specify how to retry about smithy

[mtdowling]

Thanks for the suggestion! We really need to write up a doc on how clients should handle retries...

We have some design documents floating internally within AWS describing different approaches to retry strategies based on whether an error is a throttle vs if it's some other kind of retry. This is captured today in the @retryable trait through the throttling property. From my understanding, we were planning on only needing two strategies: one using a token bucket and AIMD, and another using exponential backoff. We'll followup here when there's more information.