Retry failed workflow when scheduling is stuck

[fazpu]

Summary

• keep the normal workflow eligibility pass unchanged
• add a narrow fallback that retries the latest failed workflow only when no normal workflow is eligible
• bypass only run_every for that same failed workflow; all other gates still apply
• add regression coverage for inner success -> outer failure -> outer retry, and for not preempting normal eligibility

Verification

• uv run ruff format src/loopy_loop/scheduler.py src/tests/test_scheduler.py src/tests/test_must_follow_success.py
• uv run ruff check src/loopy_loop/scheduler.py src/tests/test_scheduler.py src/tests/test_must_follow_success.py src/tests/test_coordinator_app.py
• uv run pytest src/tests/test_scheduler.py src/tests/test_must_follow_success.py src/tests/test_coordinator_app.py
• uv run pytest
• uv run pyright

Note: one parallel uv formatting invocation hit a transient shared .venv install race; rerunning formatting serially passed.