Decorated interiors — pass 1 (#28, #29)

[Dibujaron]

First pass of decorated interiors: interiors can now carry per-tile decoration + colour end to end, and cargo pallets derive breakbulk capacity. Closes the pass-1 scope of #28 and #29; neighbour-aware / rule-bearing tiles are deferred to #36.

The glyph vocabulary (r/e/d/p/w + colors.json) already landed on main to unblock hull authoring; this PR is the pipeline + rendering that makes it live.

What's here

• One grid of Cell. DeckGrid goes from parallel tiles/edges lists to a single grid of Cell(tile, edges, decor, color) (composite.gleam already worked in cell units, so this simplified it). Walkability/collision logic is untouched — field access only — so client-side movement prediction stays bit-for-bit identical.
• Decor + NE-corner colour, lossless round-trip. Centre decoration glyph + a colour digit (0–f) in the tile's NE corner parse into the cell and re-emit through deck_to_rows. The client re-parses the same rows, so no new wire fields.
• 16-colour palette on the wire. server/colors.json (Minecraft dye order) is forwarded on welcome like the glyph registry; sprites are greyscale and multiply-tinted by the slot colour, so retuning a hex recolours every tile using that slot with zero re-authoring.
• Derived breakbulk capacity. A hull's capacity is the count of its p cargo-pallet tiles (map-is-truth, like consoles/berths); the authored number is the zero-pallet fallback.
• Client render pass. interior_view.gd draws decor tinted; a placeholder swatch shows authored decor+colour now, before any sprite art exists. Wall fixtures (w window, v screen) draw their sprite with a plain-wall fallback.
• Decorated Mockingbird adopted as the reference test ship (26 seats, 17 beds, windows, 60 pallets → derived capacity 60).

Verification

• 240 Gleam unit tests pass (incl. the Cell refactor's parity, decor/colour round-trip, composite decor-survives-mooring, palette, derived capacity, sim helm/collision/spawn); gleam format clean.
• Client: headless parse & palette probes pass; a non-headless render capture confirmed tinted decor swatches render on the right tiles.
• Integration: harness/test_m2_interior.py has 4 failures, but the identical 4 fail on the base commit (stale M2 harness vs the M3/M3.5 station-composite evolution) — this branch adds zero new failures.
• Built subagent-driven: every task got an independent spec+quality review; a final whole-branch review (which caught two wiring bugs, both fixed here + regression-tested).

Notes for review

• Front bow door (open design Q): the 4 Q tiles on the Lower deck currently model the "big front door" as 4 docking ports. They validate and break nothing, but Q semantically means a side-on mooring airlock (the Mockingbird moors on its west flank via the Mezzanine Qs). I'd model a wide bow door as a run of = door edges across the bow wall of the hold instead — no extra mooring ports. Left as-is pending your call; happy to convert.
• Authored capacity 40 vs derived 60: the hull still authors "capacity": 40, now overridden by the 60 pallets. Left your value untouched — draw fewer pallets if you want less capacity, or I can sync the number.
• Deferred Minors (naming/doc/perf nits from the reviews) are logged for the Aesthetic interior tiles (pass 2): neighbor-aware & rule-bearing tiles #36 follow-up: tiles_edges var name, a let assert in tile_block, a double glyphs.center lookup, pallet_count fold shape, a palette decode split, _fixture_tex per-edge _deck().

🤖 Generated with Claude Code