feat(sim,render,save): finite food piles with runtime respawn (closes #112)

[LightAxe]

Summary

Closes #112. Food piles previously existed as infinite static markers seeded at scenario start. They now drain by pickup-charges and a runtime spawner places fresh piles on a time-based cadence — turning food into a strategic survival mechanic where colonies must redirect foraging when nearby sources deplete.

• Sim: `FoodPile` gains `pickupsRemaining`/`pickupsInitial` (charge counter, decoupled from carry quantity); `antPickupFood` drains charges; `tickForagerActions` splices depleted piles + records depletion; new tick step 16d `tickFoodPileSpawn` runs every 1800 ticks (~90s) with rejection-sampled grass-weighted placement that avoids colonies, entrances, rally points, existing piles, and recently-depleted neighbourhoods.
• Render: 4-bucket shrink keyed on `pickupsRemaining/pickupsInitial` so both small ephemeral piles and large strategic anchors visibly drain across their lifetime.
• Save format: bumped to v3 (`SAVE_KEY = 'subterrans:save:v3'`); v2 saves rejected via the existing `SaveVersionMismatchError` flow per the "intentional for beta" policy. New validator coverage on charge fields and the new `recentlyDepletedFood` array.

Design lock and rationale lives at #112 (planning thread, not duplicated here).

Determinism / boundary

• Scenario seeding's per-pile pickup-charge draw uses a hash of `(terrainSeed, foodPileId)` instead of advancing the world RNG — preserving byte-identical replay for every existing test that calls `createScenario(seed)`.
• Runtime spawn uses the seeded world RNG with unconditional draws so the rejection-sampling sequence is replay-stable across re-runs.
• No `Math.random` / `Date` / `performance.now` / floats / cross-boundary imports introduced. ESLint pre-existing baseline (6 errors) is unchanged.

Defaults flagged for retune

Per the issue planning thread, these defaults are best-guesses pending playtest:

• `FOOD_PILE_INITIAL_PICKUPS_MIN/MAX = 20 / 150` (pile-size variation)
• `FOOD_PILE_PICKUP_DRAIN = 1` (one charge per pickup)
• `FOOD_PILE_SPAWN_INTERVAL_TICKS = 1800` (~90s at 20Hz)
• `FOOD_PILE_SOFT_CEILING = 30` (= 2× FOOD_PILE_COUNT, well under the v109 hard cap of 60)
• `FOOD_PILE_TERRAIN_GRASS_WEIGHT = 4` vs `OTHER = 1`

Test plan

• 1936 vitest tests pass (1901 → 1936; +35 new). Includes new `food-system.test.ts` (spawn gating, placement constraints, determinism, entity-ID exhaustion, tick-dispatcher integration), `pheromone/food-trail-decays-after-depletion.test.ts` (verifies the issue's "self-balancing pheromone decay" claim end-to-end), updated `save.test.ts` (direct `parseSaveFile` v2-rejection assertion + validator strictness), and `types.test.ts` `copyWorldState` round-trip coverage for the new field.
• Three internal code-review passes — final pass clean (only one NIT: off-cycle gating test uses literal tick values that would need updating if `FOOD_PILE_SPAWN_INTERVAL_TICKS` is retuned to certain values; not blocking).
• `npm run typecheck` clean. `npm run lint` shows only pre-existing baseline errors (none introduced by this PR). `scripts/check-sim-boundary.sh` and `scripts/check-asset-paths.sh` clean.
• UAT focus — this is a sim-behavior-changing PR. Worth playtesting:
  • Does the depletion + respawn cadence feel right? Initial defaults may need retuning per playtest feel.
  • Verify the visible shrink-buckets read clearly at the current TILE_SIZE_PX.
  • Verify v2 saves boot fresh (intentional rejection — surface in any UI message you have for "no save found").

Sim version

No `LATEST_SIM_VERSION` bump. Save-format bump rejects all v2 saves outright, so every save the engine ever loads is post-spawn-mechanic — no surface where old data needs to replay under "no spawn" behaviour gates.

🤖 Generated with Claude Code

[LightAxe]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Keep it up!

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".