Retry 429 errors with exponential backoff

[jwadolowski]

As mentioned in #35 it's quite easy to lock yourself out and end up with permanently broken terraform plan / terraform apply. After a bit of thinking I came to a conclusion that it'd be great to introduce some retries when request/minute threshold is hit. Even slow-ish plan/apply is much better than current behaviour.

Current state

1. Terraform immediately fails when req/minute threshold is hit

Expected state

1. Terraform should respect the API limit and keep retrying in the background with exponential backoff

Noteworthy considerations

• it'd be great if the API informs the user about remaining requests (i.e. in form of x-ratelimit-* HTTP headers; here's how GitHub and OpenAI APIs implement that). As far as I can see, there's no such a thing at the time of writing
• does it make sense to track "in-flight" requests on the client (provider) side? Is this even feasible? Would it be reliable?
• a feature flag (part of provider block?) that enables/disables the retry logic (opt-out? It's hard for me to imagine that'd be the case, but I bet that some users may prefer current behaviour)
• customizable timeouts and deadline exceeded errors (docs)