Fix add-on so the Home Assistant Supervisor can build and run it

[jgsaez9]

Problem

Installing the add-on from this repository fails. The Supervisor builds the add-on directly from speaker_recognition_addon/, but that build context does not contain the sources the Dockerfile copies (they are only synced in by CI via sync-addon-sources.sh):

COPY pyproject.toml ./          -> "/pyproject.toml": not found
COPY speaker_recognition ./...  -> "/speaker_recognition": not found

And even with the sources present, the Alpine base image ships Python 3.12, while the server pins resemblyzer / torch to python_version < '3.10', so those ML dependencies are skipped and the server cannot run.

Fix

Base the add-on on the already-published server image (built FROM python:3.9-slim), which has the working server + ML stack baked in. The add-on Dockerfile now just adds a tiny POSIX run.sh entrypoint that maps the add-on options (/data/options.json) to the server env vars and launches it.

• Rewrite speaker_recognition_addon/Dockerfile to FROM ghcr.io/eulemitkeule/speaker-recognition:1.0.12
• Add speaker_recognition_addon/run.sh (replaces the s6/bashio service; no bashio/jq needed)
• Remove the now-unused s6 rootfs
• Add .github/workflows/validate-addon.yml to build the add-on image and smoke-test the import on every push/PR, so a broken add-on build is caught in CI
• Add .gitattributes to keep *.sh at LF (CRLF would break the shebang)

Notes

• FROM is hardcoded on purpose so the injected BUILD_FROM build-arg is ignored and the Python 3.9 base is always used.
• Arch kept at amd64 (matches the current config and the published server image).