chore(ci): migrate to scitex-ai/.github reusable workflows (pilot follow-up)

[ywatanabe1989]

What

Pilot org-level reusable-workflow migration (operator-approved 2026-07-10 incident remediation), following the pattern merged in scitex-ai/scitex-stats#69. Replaces job bodies with thin workflow_call stubs into scitex-ai/.github reusable workflows where — and only where — the local job is a clean, verified mechanical match. Triggers/permissions preserved byte-identical on the one file touched.

Converted

• auto-merge-to-develop.yaml → uses: scitex-ai/.github/.github/workflows/auto-merge-to-develop.yml@main. Verified the job body (the automerge step script) was byte-identical to the reusable workflow before converting — a clean mechanical extraction, no behavior change. New required-status-check context (if this job is ever added to branch protection): call / automerge.

Left AS-IS — do not cleanly match the current 6-workflow catalog

This repo's actual CI (verified by reading job logic, not filenames) has diverged substantially from the assumptions baked into the reusable-workflow catalog piloted on scitex-stats (self-hosted spartan-cpu runner, uv-based installs, simple job bodies). Forcing a conversion here would silently change behavior or point at an unverified runner, so per the task brief these are left untouched:

• pytest-matrix-on-ubuntu-py3-11-3-12-3-13.yml — runs on ubuntu-latest (not self-hosted spartan-cpu), with CPU-torch pre-pinning, peer-standalone package installs, a PS-170 umbrella-pin-freshness audit step, and a segfault-retry + JUnit-report pass/fail decoupling scheme (the ecosystem's C-extension stack segfaults non-deterministically at interpreter shutdown/import). None of that exists in the reusable pytest-matrix.yml. Converting would drop safety-critical logic and could target a self-hosted runner label this repo may not have.
• import-smoke-on-ubuntu-py3-12.yml — ubuntu-latest, plain pip/venv, hardcoded import scitex, vs. the reusable workflow's self-hosted/uv/derived-package-name approach.
• scitex-quality-audit-on-ubuntu-latest.yml — this audits the whole ecosystem (shallow-clones every scitex-* repo and runs scripts/quality/audit_ecosystem.py), not just the calling repo. The reusable quality-audit.yml audits only the calling repo's own pyproject.toml-derived package — different job semantics entirely, not a naming coincidence.
• rtd-sphinx-build-on-ubuntu-latest.yml — beyond building Sphinx HTML, this job also commits the built bundle back into src/scitex/_sphinx_html/ on develop pushes (feeds production doc serving). The reusable rtd-sphinx-build.yml only builds and exits. Converting would silently drop the commit-back step.
• cla.yml — has no push trigger and no owner-bypass job (simpler than the reusable cla.yml), and authenticates with a secret named CLA_PERSONAL_ACCESS_TOKEN, while the reusable workflow hardcodes secrets.GH_PERSONAL_ACCESS_TOKEN. Converting risked silently breaking the CLA-signature-branch push for non-allowlisted contributors (the env var would resolve empty). Flagging for a follow-up once the secret naming is reconciled.
• newb.yml / newb-docs-quality-on-ubuntu-latest.yml — no reusable-workflow equivalent exists yet for this doc-quality-via-LLM job (two near-duplicate files already present in this repo, independent of this migration).
• pypi-publish-and-github-release-on-tag.yml (the tag-driven release workflow) — untouched per explicit instruction: PyPI OIDC trusted publishing does not support workflow_call (invalid-publisher), and this file's own comments already document why it must stay local + SIF-based.

Branch protection

Checked gh api repos/ywatanabe1989/scitex-python/branches/{develop,main}/protection — both return 404 ("Branch protection has been disabled on this repository"). No required-status-check contexts to update.

Verification

• python3 -c "import yaml; yaml.safe_load(open(...))" passes on the one edited file.
• Diff is job-body-only; on:/permissions: blocks are byte-identical to the pre-PR file.

Do not merge

Per task instructions this PR is left for review.