Switch from vLLM nightly to stable v0.15.0

The vLLM nightly had a regression causing image processor loading
to fail. Using stable release for reliability.

Removed:
- triton-kernels git dependency (not needed for stable)
- nightly wheel index
- prerelease allow setting

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

lighton-ocr2.py

@@ -6,18 +6,11 @@
 #     "huggingface-hub",
 #     "hf-xet",
 #     "pillow",
-#     "vllm",
+#     "vllm>=0.15.0",
 #     "tqdm",
 #     "toolz",
 #     "torch",
-#     "triton-kernels @ git+https://github.com/triton-lang/triton.git@v3.5.0#subdirectory=python/triton_kernels",
 # ]
-#
-# [[tool.uv.index]]
-# url = "https://wheels.vllm.ai/nightly"
-#
-# [tool.uv]
-# prerelease = "allow"
 # ///
 
 """