What is the purpose of the change?

Follow-up to #16516 (issue #16515), as suggested by @EarthChen in the review of #16516.

• Adds BroadCastClusterInvokerTest#testFailPercentRoundingDownToZero. The failure threshold is invokers.size() * broadcast.fail.percent / 100, so with four providers and broadcast.fail.percent=10 it is 0, and the broadcast has to stop at the first failed provider, as with a percentage of 0. The test makes the second provider fail, so it also checks that the broadcast does not stop before the first failure, and asserts that the third and fourth providers are not called.
• Fixes the code comment in BroadcastClusterInvoker that calls the parameter broadcast.fail.threshold. The key is broadcast.fail.percent, and has been since it was added in fix the bug of broadcast cluster invoke #7174.

No behavior change.

Verification

• The new test passes on 3.3. With the Stop broadcast at the first failure when broadcast.fail.percent is 0 #16516 change reverted, it fails at assertFalse(invoker3.isInvoked()).
• mvn test -pl dubbo-cluster: 308 tests, 0 failures (13 skipped, as on the base branch); spotless and checkstyle are clean.

Checklist

• Make sure there is a GitHub_issue field for the change.
• Write a pull request description that is detailed enough to understand what the pull request does, how, and why.
• Write necessary unit-test to verify your logic correction. If the new feature or significant change is committed, please remember to add sample in dubbo samples project.
• Make sure gitHub actions can pass. Why the workflow is failing and how to fix it?