PropagatorBasedSpanLinksExtractor used the caller-provided parentContext
as the fallback for TextMapPropagator.extract, which meant that when a
message carried no propagation headers the extractor would link the new
span to the ambient parent (e.g. a CONSUMER receive span). Use
Context.root() internally so that missing headers result in no link
being added, and remove the now-redundant Context.root() overrides and
stale comments in the batch extractors for kafka-clients, kafka-connect
and pulsar. Document the invariant on SpanLinksExtractor.