Context propagation requirements for messaging semantic conventions

[pyohannes]

This PR is a result of discussions in the messaging workgroup and has been split off #192 for easier review.

The existing messaging semantic conventions for tracing implicitly impose certain requirements on context propagation mechanisms used: while context is usually propagated on a per-request basis, the messaging semantic conventions assume that context is propagated on a per-message basis.

This OTEP proposes an addition to the existing messaging semantic conventions to make this requirement explicit. This is very important for reasoning about design questions regarding trace and span structure. Furthermore, explicating those requirements is of crucial importance for coming up with standardized context propagation mechanisms for messaging protocols. Some drafts for such standardization exist (AMQP, MQTT, CloudEvents), and the changes in this PR should help to verify those drafts against the requirements of OpenTelemetry semantic conventions.

The addition proposed in this document does not break or invalidate any of the existing semantic conventions for messaging, but rather makes an implicit requirement explicit.

[joaopgrassi]

Overall this makes sense to me!. I left a few minor things, mostly around wording and structure of text. Feel free to address them or not.

[blumamir]

LGTM
Thank you Johannes

[tedsuo]

Thank you so much for putting this OTEP together. This is an accurate representation of the model put together by the Messaging SIG.

[carlosalberto]

There's agreement from the working group itself, please review @open-telemetry/specs-approvers

cc @yurishkuro

[carlosalberto]

@open-telemetry/specs-approvers Please review. As previously mentioned, the messaging working group approved this OTEP already, we only need one more Spec approval.