[V1] LoRA - Add triton kernels for V1

[varun-sundar-rabindranath]

Add shrink and expand triton kernels for V1.

Why do we need a new set of kernels:

• V0 sorts/groups requests based on LoRA ID. The SGMV kernels take advantage of this and groups the compute within thread blocks.
• V1 doesn't group requests based on LoRA ID. The new set of kernels have information about which input tokens map to which LoRA ID and they use this information to load the appropriate input tokens. The rest of the matmul is very similar to the SGMV kernels.

Kernel Code Change:
The new kernels re-use a lot of the code from the existing SGMV kernels. The main changes are,

1. Kernel Launch Grid formulation (this was required so the kernels are CUDAGraph compatible. Note that SGMV kernels are not)
2. Loading of the input tokens (A matrix) for the matmul.
   All other kernel code is the same as the existing SGMV kernels. I refactored the code so it can be reused.

benchmark_throughput numbers:

command:

python3 benchmarks/benchmark_throughput.py --model  meta-llama/Llama-2-7b-hf --backend vllm   --dataset ./ShareGPT_V3_unfiltered_cleaned_split.json --num-prompts 500 --max-loras 4 --max-lora-rank 8  --lora-path "yard1/llama-2-7b-sql-lora-test" --enable-lora 

main + V0 : Throughput: 6.58 requests/s, 3226.96 total tokens/s, 1683.20 output tokens/s
main + V1 : Throughput: 4.78 requests/s, 2343.59 total tokens/s, 1222.43 output tokens/s
This PR V1 : Throughput: 6.74 requests/s, 3306.53 total tokens/s, 1724.70 output tokens/s

TODO : Add micro benchmark numbers

TODO:

• torch.compile support

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀