Laguna M.1 support ships in SGLang's nightly wheels (0.5.14.dev20260618+). Replaces the `git clone` + `pip install -e "python[all]"` block with an install from the nightly wheel index.
Uses the **CUDA 13** index (`/whl/cu130/`): the nightly's dependencies are pinned to the CUDA 13 stack (`flashinfer_python[cu13]==0.6.12`, `torch==2.11.0`), and the matching `sglang-kernel 0.4.4+cu130` is served there.
Supersedes #5 (which used the cu129 index).
