Check for regressions between PyPI and reqstool-client@main CI runs

[jimisola]

Context

`.github/workflows/build.yml` now runs the `reqstool status`/`reqstool validate` steps twice via a matrix (`reqstool-source: [pypi, main]`) — once against the latest released PyPI `reqstool` package, once against `reqstool-client`'s `main` branch (installed via `pip install "reqstool @ git+https://github.com/reqstool/reqstool-client.git@main"\`).

This was added so CI exercises both the stable release and the in-development `main` branch, since `reqstool-client` is intentionally holding off on its next release until the org-wide OpenSpec dogfooding rollout (see `PLAN_dog_fooding.md`) is complete — `main` has accumulated real fixes (e.g. #411) that aren't on PyPI yet.

What to check

• Periodically compare the two matrix legs' `reqstool status` output for unexpected divergence (beyond the known, intentional gap — `reqstool validate --strict` only runs on the `main` leg since that subcommand isn't published yet).
• Once `reqstool-client` cuts its next release, confirm the `pypi` leg's output matches the `main` leg's output exactly, then collapse this back down to a single non-matrixed job (drop the `main` leg, restore `reqstool validate --strict` unconditionally).
• Watch for the `pypi` leg silently masking a regression that only the `main` leg would catch (or vice versa) during the rollout period.

Related: reqstool-client#412 (CLI/MCP consolidation, found during this same rollout).