Fix shared state in retry

[sebastiangrail]

This fixes an issue where state is shared between different start invocations of the same producer.

let p = someProducer.retry(upTo: 1, interval: interval, on:scheduler)

p.start(...) // This consumes retries for the purpose of delaying errors
...
p.start(...) // Retries for the purpose of delaying errors can be negative here, resulting in the last error not being sent immediately.

Checklist

• Updated CHANGELOG.md.

[sebastiangrail]

@andersio Could you review this PR, please?

[NachoSoto]

Could you write a test for this?

[Deloww]