docs: Warn about Task lifecycle in Evaluator.evaluate

The returned Task starts executing immediately, but asyncio only
holds a weak reference to it. If the caller drops the Task before
it completes, garbage collection can cancel it mid-execution and
the judge evaluations silently never run. Document the strong-
reference contract so callers know to retain it.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: jsonbailey

packages/sdk/server-ai/src/ldai/evaluator.py

@@ -40,9 +40,9 @@ def evaluate(
         """
         Run all configured judges against the given input/output pair.
 
-        Schedules the judge evaluations as an asyncio Task and returns it
-        immediately. The caller can await the task to get results or pass it
-        to tracking helpers.
+        The returned Task starts judge evaluations immediately. The caller
+        must retain a reference to the results until it resolves to ensure
+        Judge evaluations are completed.
 
         :param input_text: The input that was provided to the AI model
         :param output_text: The AI-generated output to evaluate