Port phase 5: Vulcanus spawn, biomes and elevation (#225)

Completes the chain to vulcanus_elevation. Ports vulcanusSpawn.ts,
vulcanusBiomes.ts and vulcanusElevation.ts, and closes the
vulcanus_temperature the climate layer deferred, since it reads vulcanus_elev
and could not be ported until that existed.

Tier 1 grades fourteen more fields as frozen exact f32 match counts. Every one
was measured again on the TypeScript side against the same fixture and all
fourteen agree, so they are the distance both ports sit from the game:

  spawn, of 410        startingArea 371, startingCircle 247, ashlandsStart 61
  biomes, of 434       mountain_volcano_spots 359, mountains_raw_volcano 163,
                       mountains/ashlands/basalts_biome_full 128/107/127,
                       mountains/ashlands/basalts_biome 403/402/408
  elevation, of 434    elev 113, elevation 113, temperature 196

Three readings worth keeping.

The oracle cannot see the -500 clamp, and that was checked rather than
assumed. vulcanus_elevation is max(-500, elev), and the captured elev bottoms
out at -58.77 - so the two columns are the same field at all 434 positions,
0 of 434 differ, and a port that dropped the max would score 113 as well. Both
are graded anyway because they separate for free the day a capture reaches a
deep lake, and the clamp's real test lives in the module, where it constructs
the case the fixture does not.

The clamp flatters a count, now measurably. The three clamped biomes score
403, 402 and 408 against their own unclamped sources at 128, 107 and 127. Same
quantity, times two, clamped: nothing improved between them, the clamp
saturates at 0 or 1 over most of the map, and a saturated position is exact
for free. The *_biome_full row is the port's real score.

A discrete output scores like a discrete output. mountain_volcano_spots at
359 is the highest unclamped count in the Vulcanus port, and it is dominated
by which single candidate survives per region - a choice a sub-ULP error
almost never changes. That is the same property CLAUDE.md records for
voronoi_cell_id.

Elevation's worst residual is 1.332e-1, which is only alarming until the scale
is read: the field spans -58 to +1024 here, so it is about 1.3e-4 relative,
the same order as every layer upstream. An absolute bound would need re-tuning
per field for scale alone, which is a third argument for counting matches.

Notes on the port itself:

- vulcanus_biomes keeps a real cache, unlike every layer so far. raw_spots at
  one point reads selected spots from up to four neighbouring regions, which
  is cross-position state, so the region cache is a RefCell<BTreeMap> and eval
  stays &self. BTreeMap rather than HashMap because a determinism-critical
  port should not carry a container with unspecified iteration order, even
  where nothing iterates it yet.
- volcano_area is evaluated at every spot candidate and pulls the whole
  pre-volcano chain there. The TypeScript memoizes those; this recomputes
  them. Nothing on the render path reaches this layer yet, so it is
  correct-first by choice, and the note says to measure before that changes.
- The mountains pre-volcano split is a named method rather than an inline
  local. Collapsing the two stages would be an infinite recursion, which
  announces itself, but reading volcano_area off the POST-volcano raw would
  not.
- Three numbers in spawn are transcription traps and each has its own test:
  basalts_start's distance is a bare 250 where both its siblings scale by r,
  mountains takes half the distortion coefficient of the other two, and the
  three results carry different multipliers.
- min(a, a) in vulcanus_ashlands_func is transcribed as written. The two basis
  calls are byte-identical in the Lua, so it is the identity, and a test says
  so - collapsing it later should be deliberate.
- cliff_elevation is a separate entry point because multisample's offsets are
  in the CONSUMING program's grid units: the cliff generator's 4-tile lattice
  moves the field 16 tiles for a dx of 4, which #83 measured.

All seven Vulcanus tier-1 tests were watched going red under --features
poison. engine.wasm is unchanged and still byte-matches its source.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01X2wz8eFazisYuENEkmBvvG

Author: wormeyman

crates/fmw-noise/src/expressions/mod.rs

@@ -15,7 +15,10 @@ pub mod fulgora_scrap;
 pub mod fulgora_shared;
 pub mod fulgora_stack;
 pub mod starting_spot_at_angle;
+pub mod vulcanus_biomes;
 pub mod vulcanus_climate;
 pub mod vulcanus_cracks;
+pub mod vulcanus_elevation;
 pub mod vulcanus_helpers;
 pub mod vulcanus_seed;
+pub mod vulcanus_spawn;