[jaeger exporter] Document mapping of service name

[yurishkuro]

Changes

Documents how service name must be mapped to Jaeger spans.

[CodeBlanch]

I'm wondering if we should support Jaeger at all in the spec? It can only detect dependencies when all services in a trace report data into the same instance. Otherwise dependencies are visualized as internal. That works great academically, but shouldn't OpenTelemetry promote trace visualizations with support for real-world enterprise use cases where there will be a heterogeneous mix of instrumented reporting and non-instrumented or non-reporting instrumented services? I'm worried about the support overhead on the SIGs having to explain that and triage issues from confused or mislead users.

Reference: jaegertracing/jaeger-ui#594

[jkwatson]

If a provided Resource doesn't contain service.* attributes, and service.name in particular, how should the exporter behave? Should the exporter be configured with a fallback serviceName? Or should it just error out on trying to export those spans?

[jkwatson]

#1237