LLM: update to vllm version 0.11.0, upgrade to cuda 12.9.1

[ad-astra-video]

Two small updates in this:

1. update to vllm 0.11.0.
2. update to cuda 12.9.1 for docker image matching to vllm docker build
3. ensures tensor and pipeline a parallel check does not fail to load pipeline
4. removes max_num_batched_tokens = max_model_len to enable usage of chunked_prefill on by default

livepeer-ai-llm  | INFO 10-29 11:01:00 [model.py:1510] Using max model len 200000
livepeer-ai-llm  | INFO 10-29 11:01:00 [scheduler.py:205] Chunked prefill is enabled with max_num_batched_tokens=2048

[Copilot]

Pull Request Overview

This PR updates the LLM dependencies and Docker environment to support vLLM 0.11.0. The changes primarily involve upgrading vLLM from version 0.10.0 to 0.11.0 and updating the Docker base image to use CUDA 12.9.1 to match vLLM's requirements.

Key Changes:

• Updated vLLM dependency from 0.10.0 to 0.11.0
• Upgraded Docker base image from CUDA 12.1.1 to CUDA 12.9.1 with Ubuntu 22.04
• Modified Python package version constraints from pinned (==) to minimum versions (>=)

Reviewed Changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
runner/requirements.llm.in	Updated vLLM version constraint from 0.10.0 to 0.11.0
runner/requirements.llm.txt	Regenerated dependency lockfile (entire file removed, will be regenerated)
runner/docker/Dockerfile.llm	Updated CUDA base image to 12.9.1 and changed package version constraints to use minimum versions

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[ad-astra-video]

@mikezupper when the new image builds for this PR can you test it by chance?