[MR V2] Migrate to Model Runner V2

Squashed migration rebased onto current main. Collapses the 46-commit
dev/migrate-MR-v2 history (10 merge commits + superseded pre-rebase work)
into a single commit, so the diff reflects only the genuine MR-V2 feature
(worker_v2/ runners + model_states, omni schedulers, stage engine pipeline,
pooling_output serialization bridge). Content already merged to main
separately (step_audio2, tts_adapters, npu/_310p, etc.) drops out.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/worker_v2/test_init_model_state.py

@@ -0,0 +1,83 @@
+"""Unit tests for init_omni_model_state factory dispatch."""
+
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+import torch
+
+from vllm_omni.worker_v2.model_states import (
+    _OMNI_ARCHITECTURES,
+    init_omni_model_state,
+)
+
+pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
+
+
+def _make_vllm_config(architectures):
+    model_config = SimpleNamespace(architectures=architectures)
+    return SimpleNamespace(model_config=model_config)
+
+
+@patch("vllm_omni.worker_v2.model_states.omni_model_state.OmniModelState.__init__", return_value=None)
+def test_qwen3_omni_dispatches_to_omni_model_state(mock_init):
+    cfg = _make_vllm_config(["Qwen3OmniMoeForConditionalGeneration"])
+    model = MagicMock()
+    device = torch.device("cpu")
+
+    state = init_omni_model_state(cfg, model, None, device)
+
+    from vllm_omni.worker_v2.model_states.omni_model_state import OmniModelState
+
+    assert isinstance(state, OmniModelState)
+
+
+@patch("vllm_omni.worker_v2.model_states.omni_model_state.OmniModelState.__init__", return_value=None)
+def test_mammoth_dispatches_to_omni_model_state(mock_init):
+    cfg = _make_vllm_config(["MammothModa2ForConditionalGeneration"])
+    model = MagicMock()
+    device = torch.device("cpu")
+
+    state = init_omni_model_state(cfg, model, None, device)
+
+    from vllm_omni.worker_v2.model_states.omni_model_state import OmniModelState
+
+    assert isinstance(state, OmniModelState)
+
+
+@patch("vllm_omni.worker_v2.model_states._upstream_init_model_state")
+def test_unknown_arch_delegates_to_upstream(mock_upstream):
+    mock_upstream.return_value = MagicMock()
+    cfg = _make_vllm_config(["LlamaForCausalLM"])
+    model = MagicMock()
+    device = torch.device("cpu")
+
+    state = init_omni_model_state(cfg, model, None, device)
+
+    mock_upstream.assert_called_once_with(cfg, model, None, device)
+    assert state is mock_upstream.return_value
+
+
+def test_omni_architectures_set_contains_expected():
+    expected = {
+        "Qwen3OmniMoeForConditionalGeneration",
+        "Qwen2_5OmniForConditionalGeneration",
+        "MammothModa2ForConditionalGeneration",
+        "MiMoAudioForConditionalGeneration",
+        "MammothModa2ARForConditionalGeneration",
+        "Qwen3TTSTalkerForConditionalGeneration",
+        "Qwen3TTSCode2Wav",
+    }
+    assert _OMNI_ARCHITECTURES == expected
+
+
+@patch("vllm_omni.worker_v2.model_states._upstream_init_model_state")
+def test_none_architectures_delegates_to_upstream(mock_upstream):
+    mock_upstream.return_value = MagicMock()
+    cfg = _make_vllm_config(None)
+    model = MagicMock()
+    device = torch.device("cpu")
+
+    init_omni_model_state(cfg, model, None, device)
+
+    mock_upstream.assert_called_once()