docs(@projects/@magic-civilization): 📝 p3-18 — P4 transport core done; P5 is functionally required (UI uses pathfinder.gd)

Records transport P4a/b/c complete + the finding that the rendered game moves via
GDScript pathfinder.gd (not the Rust pathfinder), so embark works headless only
until P5 plumbs the embark gate into the GDScript path (or routes movement through
Rust). Transport refinements (carried-unit target-protection, AI-use) deferred.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

.project/objectives/p3-18-water-crossing-embark-transport.md

@@ -77,16 +77,33 @@ two-tier Civ model on the existing tree (`public/resources/techs/naval.json`):
   embark model. mc-core 263 / mc-ai 287 / mc-turn 244 green. Commit `f830a1ce0`.
   (api-gdext `legal_actions_for`/`can_invoke` keep the embark FFI inputs as
   vestigial/ABI-stable — trimmed in P5.)
-- [ ] **P4 — transport load/carry/unload.** Implement the `transport` keyword:
-  load adjacent land units (≤ capacity) onto a transport ship, they move with it,
-  unload onto adjacent land. Carried units protected. Unit tests. *(Large fresh
-  mechanic — cargo state on MapUnit + load/unload + carry-move + combat. Pure
-  Rust, cargo-test verifiable.)*
-- [ ] **P5 — GDScript pathfinder mirror + FFI trim.** `pathfinder.gd::_is_passable`
-  mirrors the embark gate (or delegates to the Rust pathfinder — Rail 1); drop the
-  vestigial embark inputs from `legal_actions_for`/`can_invoke` + their GDScript
-  callers (`unit_panel.gd`, `unit.gd`, `test_unit_actions.gd`). *(Needs the build
-  host: GUT + dylib; local cargo resolver panics on gdextension-api.)*
+- [x] **P4 — transport load/carry/unload (core).** ✅ Built per owner direction.
+  - **P4a** (`ef697f492`): catalog `UnitStats.keywords` + `is_transport()` +
+    `TRANSPORT_CAPACITY=2` (data-driven, no unit id hardcoded); `MapUnit.carrier_id`.
+  - **P4b** (`f05aaff2a`): `process_one_move` board (step onto an adjacent friendly
+    hull → `carrier_id`, capacity-gated, no embark tech needed) / carry (hull move
+    drags cargo) / disembark (carried unit → adjacent empty land). Modelled as
+    automatic moves, no new explicit actions. Tests: `transport_board_carry_then_unload`,
+    `transport_rejects_boarding_when_full`.
+  - **P4c** (`93ce7c848`): `prune_orphaned_cargo` after the PvP phase — cargo lost
+    with a destroyed hull (keeps `unit_upkeep` aligned). Test
+    `destroyed_transport_loses_cargo`.
+
+  **Refinements deferred (documented):** carried units ride the hull's hex
+  (stacked) — fully shielding them from being *individually* targeted needs combat
+  target-selection changes that touch 1UPT assumptions (rare for a tier-9 hull;
+  the prune already handles hull-death). AI does not yet load units onto transports
+  (negligible value — tier-9 hull the AI rarely builds; embark covers crossing).
+- [ ] **P5 — GDScript pathfinder mirror + FFI trim. FUNCTIONALLY REQUIRED.** The
+  rendered game moves units via `world_map.gd → PathfinderScript.find_path`
+  (`pathfinder.gd`), NOT the Rust pathfinder — so embark currently works **headless
+  only** (GdPlayerApi → Rust `process_one_move`), and a human playing the UI cannot
+  embark until this lands. Either plumb the owner's `EmbarkLevel` into
+  `pathfinder.gd::_is_passable`/`find_path`/`find_path_with_fog` + their callers,
+  or (Rail-1 preferred) route world_map movement through the Rust pathfinder. Also
+  drop the vestigial embark inputs from `legal_actions_for`/`can_invoke` + GDScript
+  callers. *(Needs the build host: the bridge must expose `embark_level` to
+  GDScript + GUT/dylib verify; local cargo resolver panics on gdextension-api.)*
 - [ ] **P6 — end-to-end.** A headless 1v1 (both sides driven) reaches a decisive
   `game_over` by crossing water to capture the enemy capital — the demo that
   motivated this objective.