[Core] Fix circular reference which leaked llm instance in local dev env (#4737)

Storing exception frame is extremely prone to circular refernece because it contains the reference to objects.

When tensorizer is not installed, it leaks llm instance because error frame has references to various modules which cause circular reference problem.

I also found spec decoding has a circular reference issue, and I solved it using weakref.proxy.

tests/basic_correctness/test_basic_correctness.py

@@ -3,9 +3,12 @@
 Run `pytest tests/basic_correctness/test_basic_correctness.py`.
 """
 import os
+import weakref
 
 import pytest
 
+from vllm import LLM
+
 MODELS = [
     "facebook/opt-125m",
     "meta-llama/Llama-2-7b-hf",
@@ -13,6 +16,16 @@ MODELS = [
 VLLM_ATTENTION_BACKEND = "VLLM_ATTENTION_BACKEND"
 
 
+def test_vllm_gc_ed():
+    """Verify vllm instance is GC'ed when it is deleted"""
+    llm = LLM("facebook/opt-125m")
+    weak_llm = weakref.ref(llm)
+    del llm
+    # If there's any circular reference to vllm, this fails
+    # because llm instance is not GC'ed.
+    assert weak_llm() is None
+
+
 @pytest.mark.parametrize("model", MODELS)
 @pytest.mark.parametrize("dtype", ["half"])
 @pytest.mark.parametrize("max_tokens", [5])