feat(simulation): T-1170 A1 — river_downstream pointers; pole-edge drains are not mouths

RiverNetwork gains river_downstream: Vec<u8> (serde-default, parallel
to river_cells): values 0-7 index drainage::D8 (row,col deltas, N/S/E/
W/NE/NW/SE/SW order); sentinels MOUTH=8, EDGE_DRAIN=9, TERMINAL=10
(reserved — the future endorheic-basin hook, Ruling 2c/7b). Captured
in extract_river_network's existing pass (fdir already in scope, one
map, no new grid pass). Pole-edge D8 exits reclassify as EDGE_DRAIN
and leave the mouths list (Ruling 3f); flat-peak interior no-outflow
cells get EDGE_DRAIN too. cascade_golden deliberately re-pinned: GJ1c
mouths 19->3 — sixteen were pole-edge artifacts, exactly Jeroen's
'circles with no sea in sight'; river_cells/attractors counts
unchanged (pure reclassification + additive field). 21/21 drainage
tests incl. mouth-sentinel/mouths-list bijection on the real fixture
and a synthetic pole-draining-grid case.

Tickets: T-1170

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

server/tests/cascade_golden.rs

@@ -18,6 +18,18 @@
 //! Golden captured on x86_64. The downsample and sub-biome cost use f32, so a
 //! different architecture could in principle round differently — regenerate
 //! per-arch if CI ever moves off x86_64.
+//!
+//! **Deliberate re-pin (T-1170 Ruling 2a/2c/3f, A1):** `RiverNetwork` gained
+//! an additive `river_downstream: Vec<u8>` field (per-`river_cells`-entry D8
+//! downstream pointer + MOUTH/EDGE_DRAIN sentinel), and `extract_river_network`
+//! stopped classifying pole-edge D8 exits (flow running off the grid's
+//! top/bottom row) as `mouths` — they are grid artifacts, not river-meets-sea
+//! events (Ruling 3f, Jeroen's capture question). On this fixture (GJ1c,
+//! 256×128 downsample) that drops `mouths` from 19 to 3: 16 of the 19 were
+//! pole-edge exits (now `RIVER_DOWNSTREAM_EDGE_DRAIN`), leaving the 3 real
+//! sea-adjacent mouths (`RIVER_DOWNSTREAM_MOUTH`). `river_cells`/`attractors`
+//! counts are unchanged (93/256) — this is a pure re-classification + one new
+//! additive field, not a drainage-algorithm change.
 
 use std::path::PathBuf;