perf(render): reuse stable session layouts

[TypeThe0ry]

What changed

• Adds a thread-safe, single-entry exact layout cache per table session.
• Keys every field read by layout generation, including center, variant, wall/dora/open-door state, seat hand selections, discards, melds, riichi, and sticks.
• Reuses immutable center, unchanged seat plans, and wall/dora components on partial changes.
• Keeps the original cold precompute implementation intact.
• Adds dependency-by-dependency cache invalidation tests and JMH scenarios.

Why

Stable refreshes rebuilt the same immutable layout objects, allocating roughly 31–36 KB per operation.

Impact

The cache is internal to a table session. Layout values, entity placement, public APIs, and rendering behavior remain unchanged.

Evidence

• stable hit: about 186 ns/op with effectively zero allocation
• seat-only miss: about 1.85 µs and 17.9 KB/op
• wall-only miss: about 1.56 µs and 9.6 KB/op
• eight-pair cold-path guardrail found no statistically significant regression

Validation

• SessionRenderLayoutCacheTest
• jmhJar
• spotlessCheck
• full integration test suite passed before commit splitting

[Arbousier1]

Formal trusted A/B passed with PASS_OPTIMIZED. Evidence: https://github.com/EllanServer/MahjongEngine/actions/runs/29645452299.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.