Differentiate between active/passive message consumption when translating otel spans

[stuartnelson3]

Currently, incoming otel spans do not have the required metadata to differentiate between whether the message is being consumed actively or passively. The consumption style (active/passive) decides whether we should create a new transaction or a new span, according to our messaging spec.

Because we cannot differentiate between the two, we have decided to always create new transactions when translating incoming otel spans with type=CONSUMER, making the assumption that the majority of message consumption will be passive.

There is a long-standing specification issue discussing adding the information to make this differentiation possible.

Commit adding an explanatory comment above the relevant code:
98d4f85

[axw]

There's also a pre-OTel issue: census-instrumentation/opencensus-specs#229. We might want to push for recording the "local root ID" in spans, which would enable us to address this issue as well as set transaction.id on spans like we do for spans produced by Elastic APM agents.

[cyrille-leclerc]

Just discovering this limitation, the solution to create a transaction looks very wise to me, I feel it makes this improvement a lower priority.