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/docs/deployment/integrations/kserve.md

@@ -0,0 +1,5 @@
+# KServe
+
+vLLM can be deployed with [KServe](https://github.com/kserve/kserve) on Kubernetes for highly scalable distributed model serving.
+
+You can use vLLM with KServe's [Hugging Face serving runtime](https://kserve.github.io/website/docs/model-serving/generative-inference/overview) or via [`LLMInferenceService` that uses llm-d](https://kserve.github.io/website/docs/model-serving/generative-inference/llmisvc/llmisvc-overview).