[fix][test] Fix flaky PublishRateLimiterOverconsumingTest

[BewareMyPower]

fixes #26006

1. Test rewritten from consumer-driven to producer-driven rate measurement
2. Uses totalMessages - getCapacity() to exclude initial burst, then asserts effective refill rate ≤ 1.4x configured rate
3. One-sided assertion (upper bound only) — won't detect under-delivery
4. Silent pass if capacity ≥ totalMessages (not possible with current params: 3000 msgs, ~500 capacity)
5. No consumers created — intentional simplification

[lhotari]

The flakiness could have been introduced by Netty 4.2 migration. Checking into this too since it could also be a production code issue.

[lhotari]

#26013 fixes the actual root cause which is a regression in Netty.

I agree that the test can be improved since there are some flaws. Some of the behavior is intentional such as testing batch messages.

One of the main changes in this PR is that it replaces producer.sendAsync with producer.send. When synchronous send is used, there will be only one message in flight for each producer at a time, which is a different scenario. Batching cannot be tested with send. It feels that both cases (send and sendAsync) should be tested as well as batching.
The way how timing is calculated, is way better in this PR. For sendAsync, it could be handled when the future completes.