Fix the backoff bounds in a retry helper

Your service retries a flaky downstream call with exponential backoff and full jitter: each sleep is a random time from 0 up to a bound, and the bound grows after every attempt. A reviewer asks you to defend the schedule, so you want a function that returns just the bounds.

An assistant wrote backoff_delay_bounds(base_ms, factor, max_delay_ms, attempts, context_cancelled). It is in the editor. It has a bug. Find it and fix it.

Input and output:

• base_ms: the first bound, in milliseconds.
• factor: what each bound is multiplied by to get the next.
• max_delay_ms: the cap. No bound is ever larger.
• attempts: how many calls the loop makes. There is one sleep between each pair of calls.
• context_cancelled: true when the request was cancelled before the loop started.
• Return the list of bounds, in milliseconds.

Rules the tests check:

• If context_cancelled is true, return an empty list. Check this first.
• If attempts is 1 or less, return an empty list.
• A negative base_ms counts as 0, a factor below 1 counts as 1, and a negative max_delay_ms counts as 0.
• Return attempts - 1 bounds. The first is base_ms and each next one is the previous bound times factor. Every bound, the first included, is capped at max_delay_ms.