NonBalancingIntegrationSpecPekkoHttp: Using pick-first should send requests to a single endpoint that is restarted in the middle

```
[info] NonBalancingIntegrationSpecPekkoHttp:
[info] Using pick-first (non load balanced clients) - pekko-http
[info] - should send requests to a single endpoint (169 milliseconds)
[info] - should send requests to a single endpoint that is restarted in the middle *** FAILED *** (131 milliseconds)
[info] 50 was not equal to 100 (NonBalancingIntegrationSpec.scala:103)
[info] org.scalatest.exceptions.TestFailedException:
[info] at org.scalatest.matchers.MatchersHelper$.indicateFailure(MatchersHelper.scala:397)
[info] at org.scalatest.matchers.should.Matchers$ShouldMethodHelperClass.shouldMatcher(Matchers.scala:7299)
[info] at org.scalatest.matchers.should.Matchers$AnyShouldWrapper.should(Matchers.scala:7347)
[info] at org.apache.pekko.grpc.scaladsl.NonBalancingIntegrationSpec.$anonfun$new$4(NonBalancingIntegrationSpec.scala:103)
```

Start with NonBalancingIntegrationSpec.scala around line 103 and run the failing NonBalancingIntegrationSpecPekkoHttp test. Trace the pick-first scenario where the endpoint is restarted in the middle, then verify that the requests still reach a single endpoint and the assertion reaches 100 rather than 50.

Written by the indexing model from the issue text.