make golu pip installable

[borawhocodess]

GOAL: make GoLU installable as a python package to enable faster and broader use across

summary of changes

• add golu/__init__.py as a package entrypoint.
• add pyproject.toml with basic metadata and minimal deps (torch, ninja)
• load CUDA kernel sources relative to the module instead of the repo root
• add a README snippet showing how to build and install.
• to build i used (uv build)

layout note for future

PyPA recommends a src/ layout (src/golu/) for new packages.

for this initial PR i kept the existing layout to keep the diff small, happy to follow up with a src/golu/ refactor if you prefer.

everything in this PR is meant as a rough starting point and subject to change, let's adjust or drop pieces as you prefer, let's make golu pip installable 🚀