Fix mushroom cloud being clipped by simulation grid

[FraGGerSheep]

Summary

The mushroom cloud was being cut off at the top because the MarchingCubes grid was a 12×12×12 cube centered at Y=0, so the simulation only ran from world Y=-6 to Y=+6. The explosion injection at N*0.12 mapped to world Y≈-4.5 (underground!), and as the cloud rose it hit the ceiling at Y=+6 where _setBoundary zeroed its density — literally clipping the cap.

This PR retunes the simulation volume and the cloud dynamics so the full mushroom shape — narrow stem, wide cap, headroom above — actually fits and reads as an atomic mushroom.

What changed

• Tall non-square grid: mc.scale.set(8, 20, 8) + position.set(0, 20, 0) → world Y now spans 0→40 with the grid bottom at ground level.
• Open top boundary in FluidSolver.step(): density and temp at j=N-1 copy from j=N-2 (with mild attenuation) instead of being zeroed, so gas flows out rather than getting killed at the ceiling.
• Entrainment force in _addForces: low-altitude gas gets pulled toward the central axis, producing a narrow rising stem.
• Vortex ring retuned to form higher (52→74% of grid height) and wider, so the toroidal roll creates a recognizable mushroom cap above the stem.
• Calmer rise: lower buoyancyAlpha (max 5 instead of 10) and lower initial vy (2.0 instead of 3.0) so gas no longer rockets straight into the boundary.
• Camera reframe: starts at (22, 10, 32) and tracks the cap up to world Y≈28; FOV 48°; fog thinned from 0.005 → 0.003 so the cap doesn't fade out.
• Shader height ramp updated to the new world-Y range.

Test plan

• npm run dev and watch a full detonation cycle — cap should clearly form above a visible stem and not visibly clip against the top of the simulation
• Restart button still works
• FPS is comparable to before (entrainment adds one extra branch per cell; should be negligible)

🤖 Generated with Claude Code