Tidal nerf, ten-level Time Lab, Cryo hub moves and a smarter time split

[OGMods]

Summary

This branch catches the optimizer up with two game updates and gives it a way to measure itself.

• The game's numbers. Tidal Ascendancy was cut from ×1.67 to ×1.35. Absolute Zero, Infinite Grid and Stellar Forge now have ten Time Lab levels instead of five, with the same top levels.
• Measuring the solver. A survey reports how far each island lands from its upper bound, and can compare a change against a saved baseline.
• Two solver improvements found that way: Cryo Nexus layouts keep their producers on one landmass, and a run spends less time on small islands that cannot improve.

Changes

Game data

• Tidal nerf (1df6454): tidal_ascendancy is ×1.35. A timeline keeps the figure it jumped in under, so the old ×1.67 stays selectable as tidal_ascendancy_legacy, a switch on the Tidal card. Share codes give it its own byte.
• Each anomaly's figures written once (1e7d378): card text, comments, docs and tests follow the anomaly table instead of repeating the number, so the next rebalance is a one-line edit.
• Ten-level research (bb4216c): the new tables from the extractor. The building roster and anomalies are unchanged. The re-extracted icons and sprite sheet differ only by compression noise (checked pixel by pixel and side by side), so they are not included. A saved level is not migrated: the game re-rates research at once, so level 5 now reads as the new level 5, as it does in game. The Time Lab buttons wrap so ten of them fit on a phone.

Measuring the solver

• Gap survey (440da52): npm run survey solves every map under every anomaly, with obstacles cleared, and ranks islands by how much of their map's bound they leave on the table, with the run-to-run spread beside it.
• Before-and-after mode (6d340e4): --compare base.json sets every island against a baseline saved with --json and exits 1 if any got worse. Each island is judged on the mean of its runs, against a noise band set by the wider run spread. The golden fixtures cannot see what a change does to a real timed run; this can.

What the survey found, at 15s per map: the solver lands at 96–99.6% of the bound under the base rules and Tidal, and at the bound under Cryo Nexus. Singularity Isolation sits lower (93–96%), and more time barely moves it. docs/SOLVER.md already notes that the Singularity bound is loose, so most of that gap is probably the bound rather than the search.

Solver

• Cryo Nexus moves whole hubs (174fd0a): a player reported a layout with a hub left on a small second landmass. Moving it onto the main island and filling the second landmass with coolers kept the same power, but the cooler-gathering pass moves one tile at a time and cannot move a hub. A new step moves all of a smaller landmass's producers onto the main one together, then swaps buildings on the main island until power is back. It is kept only if power, every producer and the upgrade plan hold. On the reported board it reproduces the player's layout at the same 611AC. Across the eight cleared maps, 7 now keep every producer on one landmass. Shadowspire's 29-producer second landmass does not fit. It has its own 1s limit, sized for phones; a 15-producer hub takes about 170ms on a desktop.
• Time split by tile count squared (394df43): solving every island at ¼ to 2× its share showed that islands of 20 tiles or fewer reach the same result at ¼ as at 2×, while every landmass of 67+ tiles was still improving at 2×. islandBudgetsS now weights each island by tiles², with a floor of the smaller of 0.25s and its old share. Entropy Isles' main landmass goes from 12.1s to 13.8s of a 15s run. planSolve and the app's run-time estimate both use it, so the time shown on the run button matches the run.

App

• Show the full power figure on hover or tap (d74a71c).
• Show an imported layout's own board straight away (bd6fc19).
• Focus the code box when Import opens (216c7f5).
• Prettier formatting (70778ea).

Testing

• npm test: 869 passing. New tests cover the hub move on the reported board (it fails with the step disabled), islandBudgetsS, and the research tables. Tests that meant the top research level now ask for level 10. The research-then-anomaly order test now uses level 3's ×1.3, the lowest new level where the two orders give different results.
• npm run check, npm run knip and Prettier are clean.
• Time split, measured with --compare against the old split (15s × 5 runs, No Anomaly / Tidal / Singularity): no island got worse. Maps 5–8, the fragmented ones it targets, lean about +0.25%. That is within the ±0.8% run-to-run noise at this sample size, so the direction is real but the size is not confirmed.
• Cryo hub move: fresh solves of all eight cleared maps, old code against new, same seed. Power held or rose everywhere, and the maps with producers on more than one landmass went from 5 to 1.