docs(trace-retention): clarify span vs trace sampling order and behavior

Clarifies the evaluation order and semantics of span rate and trace rate in retention filters.

The previous wording could be interpreted as span and trace rates being applied independently.

Author: wmfrov

content/en/tracing/trace_pipeline/trace_retention.md

@@ -148,8 +148,9 @@ To create a retention filter:
 <div class="alert alert-warning">Configuring a trace rate can significantly increase your indexed spans usage.</div>
 
 For example, if you configure a retention filter to index spans from `service:my-service`:
-- Configuring a span rate of `100%` ensures that all spans matching `service:my-service` are indexed.
-- Configuring a trace rate of `50%` ensures that all spans from all traces with a span from `service:my-service` are indexed. Assuming traces have 100 spans in average and 5 spans from `service:my-service`, configuring a trace rate indexes the remaining 95 spans of the trace, for the trace rate percentage being configured.
+- Configuring a span rate of `50%` ensures that approximately 50% of traces that contain spans matching `service:my-service` are selected. For selected traces, all spans matching `service:my-service` are indexed.
+- Configuring a trace rate of `10%` ensures that 10% of the traces selected by the span rate are fully indexed. For those traces, all spans in the trace (not just those from `service:my-service`) are indexed. Assuming traces have 100 spans on average and 5 spans from `service:my-service`, configuring a trace rate indexes the remaining 95 spans of the trace for the configured percentage of selected traces.
+- Span rate is evaluated first, and trace rate is applied only to traces selected by the span rate.
 
 When you create a new filter or edit the retention rate of an existing filter, Datadog displays an estimate of the percentage change in global indexing volume.