[Test] Unify L2/L3 test layout, Buildkite steps, and test helpers (#2556)

Signed-off-by: wangyu <410167048@qq.com>
Signed-off-by: Huang, Zeyu <11222265+fhfuih@users.noreply.github.com>
Signed-off-by: wangyu <53896905+yenuo26@users.noreply.github.com>
Co-authored-by: Huang, Zeyu <11222265+fhfuih@users.noreply.github.com>

tests/model_executor/models/test_encoder_quant_config.py

@@ -7,19 +7,46 @@ For modelopt FP8/FP4/MXFP8 checkpoints the Thinker LM is the only
 quantized component.  Vision and audio encoder weights are BF16 with no
 FP8 scale tensors — passing quant_config to them causes FP8 kernels to
 run on BF16 weights, producing garbage embeddings.
+
+**Import note (pytest + ``tests/model_executor`` only):** :mod:`vllm_omni.quantization`
+:file:`__init__` pulls ``factory``/``inc`` and reproduces vLLM CustomOp double
+registration.  We load the same :file:`component_config.py` via
+:data:`importlib` under a *non-package* name so the package ``__init__`` is
+never executed for this file.
 """
 
 from __future__ import annotations
 
+import importlib.util
+import sys
+from pathlib import Path
 from unittest.mock import MagicMock
 
 import pytest
 
-from vllm_omni.quantization.component_config import (
-    PRE_QUANTIZED_METHODS,
-    ComponentQuantizationConfig,
-    resolve_encoder_quant_config,
-)
+_CC_MODULE_NAME = "_vllm_omni_tests_component_config_encoder_quant"
+
+
+def _load_component_config():
+    if _CC_MODULE_NAME in sys.modules:
+        return sys.modules[_CC_MODULE_NAME]
+    root = Path(__file__).resolve().parents[3]
+    path = root / "vllm_omni" / "quantization" / "component_config.py"
+    if not path.is_file():
+        raise FileNotFoundError(f"expected component_config at {path}")
+    spec = importlib.util.spec_from_file_location(_CC_MODULE_NAME, path)
+    if spec is None or spec.loader is None:
+        raise RuntimeError("failed to create import spec for component_config")
+    mod = importlib.util.module_from_spec(spec)
+    sys.modules[_CC_MODULE_NAME] = mod
+    spec.loader.exec_module(mod)
+    return mod
+
+
+_cc = _load_component_config()
+PRE_QUANTIZED_METHODS = _cc.PRE_QUANTIZED_METHODS
+ComponentQuantizationConfig = _cc.ComponentQuantizationConfig
+resolve_encoder_quant_config = _cc.resolve_encoder_quant_config
 
 pytestmark = [pytest.mark.core_model, pytest.mark.cpu]