Steps to Reproduce

1. Configure an OTEL exporter (e.g. Langfuse via OTEL_EXPORTER_OTLP_ENDPOINT)
2. Enable latest GenAI semantic conventions (OTEL_SEMCONV_STABILITY_OPT_IN=semconv_stability_opt_in_latest_genai)
3. Run an agent and inspect the exported spans
4. Observe that execute_event_loop_cycle spans carry gen_ai.input.messages as a span attribute but lack gen_ai.system / gen_ai.provider.name / gen_ai.operation.name

Expected Behavior

execute_event_loop_cycle spans should include the same common gen_ai.* attributes (gen_ai.system or gen_ai.provider.name, and gen_ai.operation.name) that are present on start_model_invoke_span, start_tool_call_span, start_agent_span, and start_multiagent_span spans. All four of those span types call _get_common_attributes(); start_event_loop_cycle_span does not.

Actual Behavior

start_event_loop_cycle_span() (tracer.py ~line 500) initializes its attributes dict with only event_loop.cycle_id and optional parent/custom attributes, then calls _add_event_messages() which (since #1768 / v1.29.0) promotes gen_ai.input.messages to span-level attributes when Langfuse is detected. The result is spans that carry gen_ai.-namespaced content attributes but lack the common identification attributes that downstream OTEL tooling (collectors, exporters, redaction processors) may rely on for filtering or policy decisions.

Possible Solution

Add a call to _get_common_attributes() in start_event_loop_cycle_span(), consistent with the other span start methods. An appropriate operation_name would need to be chosen (e.g. "execute_event_loop_cycle").

Related Issues

#1768