chore(release): v2026.10.12 — the voyage release

[marceld23]

Release prep for v2026.10.12 — the voyage release.

• CHANGELOG.md: the 2026.10.12 section (travel from a parked ship plays take-off → warp → landing Travel from a landed ship should play take-off → (warp) → landing instead of an instant world swap #1614, Ahmed co-author; water at the x = 0 seam comes to rest Water across the x = 0 seam never settles: fluid cells are keyed by two spellings and turn into phantom sources #2447, placed water wakes its pond Placing water into resting water does not wake the neighbours: the pond never grows to its new size #2448, water twice as fast with lava unchanged Water flows twice as fast (one fluid step every 0.125 s); lava keeps its speed #2449; an implausibly far target lock is released and logged Lock HUD: drop an implausibly far lock and log its source (the #2428 diagnosis never shipped; second crash report) #2450; the test-shard weights refreshed) plus the compare links.
• TODO.md: the two sections merged since v2026.10.11 are marked released.
• data/whatsnew.json: the DE+EN release post for 2026.10.12, exported with tools/devblog/export_whatsnew.py.
• data/locales/ko.json: ui.map.hyperjump_here read "이 항성계로 초점 이동" ("move the focus to this system") — now "이 항성계로 하이퍼점프". Found while translating the release post, which names this button; the other eleven community labels are fine.

ℹ️ Protocol stays 11, save version 3, terrain generation 23. The staged transit needs client and server on 2026.10.12. Shared changed via #2452 (a doc comment in ServerConfig), so the fleet roll redeploys worldhost on sha-679f8d9.

The twelve community-language "What's new?" files follow in a separate PR once the release post is live on the blog.

🤖 Generated with Claude Code

[marceld23]

Merging with --admin although shard 7/10 is red: in all three attempts every test passed (520/520) and only the PR-only 120 s fast-tier duration guard tripped, each time on different tests (attempt 1 GadgetTests.TerrainScanner_FindsNearbyOre_CostsEnergy_AndCoolsDown 122 s; attempt 2 FactoryClaimTests.Claim_ConsumesAccessCode_GrantsOwnerEdit_DeniesOthers_AndPersists 139 s; attempt 3 GadgetTests.FluidPump_HarvestsWaterToo_ButNothingElse 166 s, BioTankFixTests.AFinishedAnimalCross_… 133 s, WorldGenerationGoldenTests.GeneratedChunks_MatchThePinnedGoldens 125 s) — queue latency, not the tests' own work. This PR only touches CHANGELOG, TODO, whatsnew and one Korean label; the release gate on the tag runs the full tier anyway.

Side finding: shard 7 is the long pole since the weight refresh (12.3 min against 5–11 min on the others in attempt 1; #2452 also needed a shard-7 rerun).