Fix exponential backoff algorithm truncating exponential factor

[tinnywang]

The exponential backoff algorithm truncates the exponential factor when math.Pow(base, float64(attempt)) is converted to time.Duration (the type conversion is from float64 to int64).

This makes the delay between retries not increase exponentially when the base is a non-integer.
For example, the expected vs. actual results for the first five attempts of Exponential(100 * time.Millisecond, 1.2) are shown below:

attempt	expected	actual
1	120ms	100ms
2	144ms	100ms
3	172.8ms	100ms
4	207.36ms	200ms
5	248.832ms	200ms

To fix this, multiply the duration factor by the exponential factor before converting anything back to time.Duration.

[Rican7]

Oh wow, thanks for this Tiffany! Sorry that I'm just now getting to this.

Now that I think about it... oof... there's a few different places where gross conversions between float and int are happening... I think I'm going to audit all of those and re-think some things before merging this so that the change can be more consistent throughout the codebase.

Thanks for bringing this to my attention and for proposing these changes!