Group world data into per-category chunks, and add an itch.io build

[thatoneguy10001]

Cuts the Pages build from 400+ JS chunks to 23, and adds a build target that works on itch.io.

Chunking

engine-wasm.ts globs world JSON lazily, which emits one chunk per file — Iron & Blood alone is 154 quests, 72 NPCs and 71 rooms, so a first load fired ~400 sub-10KB requests.

Grouped by (world, category) in manualChunks. The granularity follows how the data is actually consumed:

• initEngine awaits all six categories for one world in a single Promise.all, so a category is always fetched whole or not at all — finer than category buys nothing.
• Category rather than whole-world, because listPlayableClasses runs earlier on the character-creation screen. Collapsing a world into one chunk would make class select drag 154 quests with it.
• world.json is left unmatched deliberately — it is globbed eagerly for the world-select screen, so it belongs in the entry bundle.

	Before	After
JS chunks	400+	23
Requests to reach first room	~400	17

Per-world laziness verified intact: starting Iron & Blood fetches 7 chunks and zero Millbrook chunks. A world added later gets its own chunks with no config change.

The world-iron-and-blood-npcs chunk (563 KB raw / ~186 KB gzipped) trips Vite's 500 KB warning. Expected — initEngine needs every NPC anyway, so splitting it would just restore the requests this removes.

itch.io build

npm run build:itch → tsc && vite build --base=./ --outDir dist-itch.

itch serves HTML5 games from a randomised subdirectory, so the absolute /chronos/assets/… paths the Pages deploy emits 404 there. --base=./ makes them relative. Separate outDir so it cannot collide with the Pages build; both gitignored. The Pages deploy is untouched — the workflow still supplies VITE_BASE=/chronos/.

New docs/publishing-itch.md covers build → package → upload.

Roadmap reconciliation

docs/expansion-planning.md was written 2026-06-18, before the sprint that implemented most of it, and read as though nothing had shipped — because the implementation renamed nearly everything (ActiveLoadout→PayloadSlots, VisitRoom→ReachRoom, SPD/ACC/CRIT/DODGE→agility/hit/crit_chance/evasion, CraftWeapon→AssembledWeapon).

All 27 tracked items re-verified against the code and annotated in place. 9 of 12 engine and 7 of 7 ungated content items have shipped. Genuinely outstanding: E-2 (rarity), E-3 (faction standing — a design decision, not work: no faction system exists in engine or world JSON), E-8 (SCAMP, routed around and non-blocking), and FetchItem from E-10.

Verification

• tsc --noEmit clean; 77 tests pass
• Built, served and played: world select → Iron & Blood → Vanguard → Gate of Fort Iron, 135/135 HP, items and NPC present
• Zero console errors, all requests 200

🤖 Generated with Claude Code