perf(riichi): skip structurally impossible discard reactions

[TypeThe0ry]

What changed

• Caches normalized base-tile counts by hand version.
• Checks pon, open-kan, and legal left-player chii structure before invoking the full shanten search.
• Preserves the exact ron-only result when no structural call is possible.
• Adds fixed-seed differential coverage against the original reaction path.

Why

Discard reactions were spending most CPU and allocation in furoChanceShanten even when the hand could not possibly pon, kan, or chii the discarded tile.

Impact

External APIs, event order, action availability, and ron behavior are unchanged. The fast path only skips a search after proving that no structural meld action exists.

Evidence

Paired JMH:

• impossible-call path: about 98.8% lower time and effectively zero allocation
• ron-only path: about 98.7% lower time and 99.4% lower allocation
• possible-pon path remains within the allocation guardrail

Validation

• RiichiReactionStructuralGateTest
• spotlessCheck
• full integration test suite passed before commit splitting

[Arbousier1]

Formal trusted A/B run did not pass: INCONCLUSIVE_NOISE_OR_NO_GAIN. Evidence: https://github.com/EllanServer/MahjongEngine/actions/runs/29645449549. Closing per the performance gate policy.