ComfyUI Base 2.8.0: the owned-box recipe says what a Blackwell card actually needs

hosts/local/README.md now sets out the chain the whole host hangs on, with the file each link lives in, so a
refusal can be read instead of guessed at: compute capability 12.0 becomes sm_120; sm_120 means the CUDA 13
wheels are the only stable ones; CUDA 13 means driver >= 580; and torch is therefore the newest +cu130 wheel
for the interpreter's tag rather than a pinned number.

Two practical consequences are spelled out, because both look like breakage and neither is. A driver below
580 stops the run at the gate with nothing downloaded and the venv untouched, which is the gate working. And
SageAttention is built from source here because no PyPI wheel carries sm_100/sm_120 kernels, so the build
wants the GPU visible; SAGE_ARCHS or SAGE_WHEEL is the answer when it is not.

Also the MCP shape for a box with no pod and no ssh alias, since 2.7.0's podctl mcp has nothing to point at
locally: the client runs the venv's comfy-mcp directly.

The status row stays honest. This is documentation, exercised on no hardware: the host is still a recipe and
the host-report template is there for the first person to run it end to end.

---

Step one on a fresh machine:

python3 -m zipfile -e comfyui-base.zip . && bash comfyui-base/comfyui-base-script.sh