[DEFER] Re-evaluate Windows Wine smoke cadence for CI cost/coverage balance

[mark-e-deyoung]

Summary

Windows Wine smoke tests currently run on main pushes only (not every PR) after CI fast-tiering.

Problem

Further reducing frequency (for example release-only or scheduled-only) could reduce CI cost, but risks slower detection of Windows/runtime regressions.

Proposal

Evaluate and choose one policy:

• keep current main push cadence (baseline)
• move to release + scheduled cadence
• add selective trigger based on Windows/runtime-impacting path changes

Acceptance criteria

• Chosen cadence documented in policy and workflow comments.
• Workflow reflects chosen cadence.
• Regression-detection tradeoff explicitly recorded.

Decision context

Deferred to preserve current coverage until more reliability and cost data is collected.