implement RetryPolicy for ExponentialBackoffTimed

[bbaldino]

This PR seeks to implement the changes discussed in TrueLayer/reqwest-middleware#109. Mainly: that there is not currently a way to define a RetryPolicy which can observe a maximum retry duration. There did exist ExponentialBackoffTimed which, from what I can tell, sought to implement this, but since it did not implement RetryPolicy there was no way to pass it and have each new request have an independent start time. I believe this also obsolesces the need for ExponentialBackoffWithStart, so I removed it.

I did consider another approach:

1. Rather than changing the should_retry API, we could keep the task start time in the RetryPolicy itself, but:
   • The soonest we'd be able to set the task start time would be in the first call to should_retry, which isn't really accurate
   • This makes the RetryPolicy implementation stateful, which I don't think makes sense considering only a single one is passed in.
   • We could change the policy passed in to be more of a policy "factory" or add additional methods to RetryPolicy (like request_started, but those are even bigger changes.

Definitely open to other suggestions, though I think the code here is consistent with what was discussed in the linked ticket. Will link the corresponding reqwest_middleware PR as well. corresponding reqwest-middleware PR.

[ThomWright]

So this looks good to me.

For context, I believe I made the change this way to avoid a big breaking change in the primary API of the crate: the should_retry function.

In hindsight, that might not have been the best trade-off!

Thank you for the issue and PR 🙏

I've asked @tl-rodrigo-gryzinski to take a look too.

[bbaldino]

@ThomWright we good to merge here? if so I can update the reqwest-middleware pr once we get a new retry-policies version released with this change

[ThomWright]

@bbaldino Ethan has kindly merged and published the changes.

Please feel free to make the changes in request-middleware and we'll take a look 🙂