[codex] Fix bridge feedback regressions

[casperpeters]

Summary

• Prevent responder from raising a completed major game to 5M after 1M-3M-4M, while preserving Blackwood only for agreed-trump slam-zone hands.
• Fix declarer-side card play so avoidLongHandRuff and cash-winner plan priorities do not overrule winning the current trick with a cheap trump.
• Add a visible-dummy defensive lead heuristic so a defender avoids leading an unsupported honor under dummy's higher honor mid-hand.

Validation

• npm run test:unit passed with 375 tests.
• Feedback sheet rows 3 through 6 were updated to Geimplementeerd during the implementation runs.

Notes

gh is not installed in this environment, so this PR was opened through the GitHub connector after pushing the branch with git.