test: xfail legacy flaky test (#1028)

`tests/integration/test_scenarios.py::test_scenario_initial_task_types[new_task-streaming-legacy]`
is flaky, given that it's legacy and is not used - `xfail` it. This file
is currently used as a live documentation of legacy and new executor
behavior differences and bug fixes, so there is some value in keeping
this specific scenario for both legacy and new in case we want to
compare this behavior from some other aspect.

Example failures:

-
https://github.com/a2aproject/a2a-python/actions/runs/24995840443/job/73192514384
-
https://github.com/a2aproject/a2a-python/actions/runs/25041293011/job/73344930927

---------

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

Author: ishymko

tests/integration/test_scenarios.py

@@ -1933,8 +1933,21 @@ async def listen_to_end():
 )
 @pytest.mark.parametrize('initial_task_type', ['new_task', 'status_update'])
 async def test_scenario_initial_task_types(
-    use_legacy, streaming, initial_task_type
+    request, use_legacy, streaming, initial_task_type
 ):
+    if use_legacy and streaming and initial_task_type == 'new_task':
+        # There is a race condition which manifests itself in flaky CI failures.
+        # Given that we don't use legacy by default anymore, xfail it.
+        # Flakiness rate is around 1 failure per 30 runs, so use strict=False
+        # to avoid failing on a passing xfail.
+        request.node.add_marker(
+            pytest.mark.xfail(
+                reason='https://github.com/a2aproject/a2a-python/issues/869',
+                strict=False,
+                raises=ValueError,
+            )
+        )
+
     started_event = asyncio.Event()
     continue_event = asyncio.Event()