Add vLLM v0.18.1 source tree with KV transfer abort fix

third_party/vllm/ now tracked in git for direct patch management.
Based on vLLM v0.18.1 release with one patch applied:

  vllm/v1/core/sched/scheduler.py:
    Replace fatal assert with graceful skip when KV transfer callback
    arrives for an already-aborted request during PD disaggregated serving.

Future vLLM modifications should be made directly in third_party/vllm/
and committed normally. The patches/ directory is kept as documentation
of what changed from upstream.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

third_party/vllm/requirements/rocm-test.txt

@@ -0,0 +1,114 @@
+# Common dependencies
+-r common.txt
+
+# Test infrastructure
+tblib==3.1.0
+pytest==8.3.5
+pytest-asyncio==0.24.0
+pytest-timeout==2.3.1
+pytest-cov==6.3.0
+pytest-forked==1.6.0
+pytest-rerunfailures==14.0
+pytest-shard==0.1.2
+
+# Async/HTTP dependencies
+anyio==4.6.2.post1
+    # via httpx, starlette
+aiohttp==3.13.3
+    # via gpt-oss
+httpx==0.27.2
+    # HTTP testing
+
+# Audio processing dependencies
+audioread==3.0.1
+    # via librosa
+cffi==1.17.1
+    # via soundfile
+decorator==5.2.1
+    # via librosa
+lazy-loader==0.4
+    # via librosa
+platformdirs==4.3.6
+    # via pooch
+pooch==1.8.2
+    # via librosa
+soundfile==0.13.1
+    # via librosa
+soxr==0.5.0.post1
+    # via librosa
+librosa==0.10.2.post1
+
+# Retrieval and search
+bm25s==0.2.13
+    # via mteb
+pystemmer==3.0.0
+    # via mteb
+
+# Multi-modal processing
+av==16.1.0
+    # required for audio_in_video tests
+blobfile==3.0.0
+    # Multi-Modal Models Test
+decord==0.6.0
+    # video processing, required by entrypoints/openai/test_video.py
+rapidfuzz==3.12.1
+
+# OpenAI compatibility and testing
+gpt-oss==0.0.8
+    # OpenAI compatibility tests
+schemathesis==3.39.15
+    # OpenAI schema test
+
+# Evaluation and benchmarking
+lm-eval[api]==0.4.11
+jiwer==4.0.0
+
+# Required for multiprocessed tests that use spawn method, Datasets and Evaluate Test
+multiprocess==0.70.16
+
+# Required for v1/metrics/test_engine_logger_apis.py
+ray[cgraph,default]>=2.48.0
+
+torchgeo==0.7.0
+    # via terratorch
+# MTEB Benchmark Test
+mteb[bm25s]>=2, <3
+
+# Utilities
+num2words==0.5.14
+    # via lm-eval
+pqdm==0.2.0
+    # via lm-eval
+
+# Required for fastsafetensors test
+fastsafetensors @ git+https://github.com/foundation-model-stack/fastsafetensors.git@d6f998a03432b2452f8de2bb5cefb5af9795d459
+# Required for suffix decoding test
+arctic-inference == 0.1.1
+# Required for Nemotron test
+open-clip-torch==2.32.0
+# Required for isaac Multi-Modal generation test
+perceptron==0.1.4
+# Required for the multi-modal models test
+timm==1.0.17
+# Required for plugins test
+albumentations==1.4.6
+# Pin transformers version
+transformers==4.57.5
+# Pin HF Hub version
+huggingface-hub==0.36.2
+# Pin Mistral Common
+mistral-common[image,audio]==1.10.0
+# Required for Prithvi tests
+terratorch==1.2.2
+# Required for Prithvi tests
+segmentation-models-pytorch==0.5.0
+# Required for Prithvi tests
+imagehash==4.3.2
+# Required for bitsandbytes quantization test
+bitsandbytes==0.49.2
+# Examples (tensorizer) tests
+tensorizer==2.10.1
+# Multi-modal models test (`allendou/FireRedASR2-LLM-vllm`)
+kaldi-native-fbank==1.22.3
+# Pinning numpy version
+numpy==2.2.6