Reactor instrumentation: do not make root context current

[lmolkova]

In some cases, reactor's fluxes can be created and subscribed for at startup, and live till application dies.
TracingSubscriber is created at subscription time and captures root context as current, but this context is likely irrelevant to the rest of the Flux lifetime.

E.g. the receiver will read and process messages from the queue:

client = QueueReceveierBuilder().build();
client
  .receiveMessages()
  .subscribe(msg -> System::out::println);

Now imagine we do something like this:

  Flux<Object> publish = Flux.create(sink -> {
    Span s = tracer.spanBuilder("inner").startSpan();
    try (Scope scope = s.makeCurrent()) {
      // this will invoke the subscriber callback below synchronously
      // and we'd expect that inner span should be active
      sink.next(42); 
    } finally {
      s.end();
    }
  });

  publish
      .take(1)
      .subscribe(n -> {
        // this code is invoked synchronously from sink.next
        // but in between there was a TracingSubscriber that made root context active
        // so without this change, this assert fails
        assertThat(Span.current().getSpanContext().isValid()).isTrue();
      });

This change only prevents the root context from becoming current. But if subscribe happens in scope of some span, there is no reliable way to know if it's the correct one. If not, applications can work it around by

• subscribe to long-running fluxes outside of any scope
• adding root trace context, to reactor context to avoid tracing subscriber to propagate anything as done in doesNotOverrideInnerCurrentSpansWithThereIsOuterCurrent test

[trask]

@lmolkova heads up, I merged main into your branch to fix CI

[trask]

nice, thx!