fix(llmobs): prevent evaluator spans from leaking into experiments index

[FouadWahabi]

Summary

• Evaluator LLM spans were incorrectly inheriting the experiment span's baggage and getting routed to the experiments index instead of the standard LLMObs index
• Added _clear_trace_context() call between task and evaluator execution in _process_and_evaluate to clear both the default tracer and LLMObs context providers
• Added test verifying that task spans get scope=experiments while evaluator spans do not

Root Cause

In _run_tasks_with_evaluators, the _process_and_evaluate coroutine runs _process_record (task) and _evaluate_record (evaluators) sequentially within the same asyncio Task. The experiment span created during task execution sets EXPERIMENT_ID_KEY as baggage on its trace context. After the experiment span finishes, the trace context may still carry this baggage when evaluator code runs, causing any LLM spans created by evaluators to inherit the scope=experiments attribute and get routed to the wrong index.

Test plan

• New test test_experiment_evaluator_spans_not_in_experiments_scope verifies task spans have scope=experiments while evaluator spans do not
• Existing experiment tests continue to pass

[cit-pr-commenter-54b7da]

Codeowners resolved as

ddtrace/llmobs/_experiment.py                                           @DataDog/ml-observability
tests/llmobs/test_experiments.py                                        @DataDog/ml-observability