Add Network Semantic Conventions project proposal

[svencowart]

This PR introduces the formal project proposal for Network Telemetry Semantic Conventions.

The goal of this project is to establish unified and comprehensive network semantic conventions for OpenTelemetry, enabling traditional network observability methods (like NetFlow, IPFIX, SNMP) to integrate natively into the ecosystem.

Key Objectives & Deliverables

• Core Conventions: Expand, clarify, and maintain the network, source, and destination namespaces.
• Traditional Telemetry: Define deep network semantic conventions for Netflow (metrics and trace spans) and SNMP (metrics and trap logs).
• Ecosystem Integration: Improve the usability of network-related collector receivers (e.g., snmpreceiver, netflowreceiver) and collaborate closely with the Semantic Conventions, Entities, System, and OBI SIGs.

[MatthieuNoirbusson]

Solid scope — good to see SNMP (metrics, MIB objects, traps) and entity definitions in here alongside the flow and namespace work.

One input from the entity/SNMP side, more as a data point than a request:

In our SNMP-discovery work, most of the operational value lands in the edges between entities rather than the entities alone — adjacencies from LLDP/CDP, an interface's VLAN membership, tunnel endpoints, route next-hops. We model those as relationships embedded on the source entity's state event, following #4836 (shipped in v1.58.0). Since #4836 makes that a first-class option now, it's probably worth deciding early whether entities like adjacencies/peers are meant as relationships or as attributes — that choice shapes a lot downstream.

The entity list currently reads entity-by-entity; if entity shape is what informs the rest of the conventions, pulling that discussion earlier could pay off. Either way, here's our candidate model as a reference for the entity-definitions work — device/interface/route/address with connected_to / has_route / bound_to: https://gist.github.com/MatthieuNoirbusson/84ed1cef498780bdf70dd507b384c1d9. It's one implementation's model, meant to complement the MIB-normalization work rather than overlap it.

[mariomac]

LGTM!

[pinoOgni]

lgtm!

[robcowart]

The plan looks good.

[lmolkova]

LGTM

[svencowart]

This project will allow us to close issue #3769 within the semconv repo.