refactor: reduce copied vLLM 0.29 compatibility code

Reuse upstream weight mapping and multimodal processing, consolidate Omni processor bridges, and trim unused Hunyuan internals. Route no-media kwargs through the active renderer and preserve GLM generation prompt formatting.

Validation: 453 CPU tests passed, 7 skipped; processor imports, Ruff, and diff checks passed. GPU and end-to-end CI not rerun.
Signed-off-by: tzhouam <tzhouam@connect.ust.hk>

tests/model_executor/models/test_weight_loader.py

@@ -4,9 +4,7 @@
 import pytest
 import torch
 from torch import nn
-from vllm.model_executor.models.utils import WeightsMapper
-
-from vllm_omni.model_executor.models.weight_loader import AutoWeightsLoader
+from vllm.model_executor.models.utils import AutoWeightsLoader, WeightsMapper
 
 pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
 
@@ -20,21 +18,18 @@ class _TinyModel(nn.Module):
         self.rotary_embed.register_parameter("inv_freq", nn.Parameter(torch.zeros(1)))
 
 
-def test_auto_weights_loader_preserves_removed_skip_arguments() -> None:
+def test_upstream_mapper_preserves_stage_exclusions_after_renaming() -> None:
     model = _TinyModel()
     model.skipped.weight.data.zero_()
 
-    loaded = AutoWeightsLoader(
-        model,
-        skip_prefixes=["skipped."],
-        skip_substrs=["rotary_embed.inv_freq"],
-    ).load_weights(
+    loaded = AutoWeightsLoader(model).load_weights(
         [
             ("checkpoint.kept.weight", torch.ones(1, 1)),
-            ("skipped.weight", torch.full((1, 1), 2.0)),
+            ("checkpoint.skipped.weight", torch.full((1, 1), 2.0)),
             ("rotary_embed.inv_freq", torch.ones(1)),
         ],
-        mapper=WeightsMapper(orig_to_new_prefix={"checkpoint.": ""}),
+        mapper=WeightsMapper(orig_to_new_prefix={"checkpoint.": ""})
+        | WeightsMapper(orig_to_new_prefix={"skipped.": None}, orig_to_new_substr={"rotary_embed.inv_freq": None}),
     )
 
     assert loaded == {"kept.weight"}