Compute CUDA timestep at step end

[AhmedSalih3d]

Motivation

• Avoid per-particle scalar GPU indexing (and the need for CUDA.@allowscalar) by computing the adaptive time step after the second neighbor loop.

Description

• Remove the per-step max_visc/min_dt_force buffers and their updates from the CUDA loop and compute the time step at the end of the iteration with Δt(Positionₙ⁺, Velocityₙ⁺, Acceleration, SimConstants, SimKernel), updating src/SPHCUDA.jl accordingly; commands executed during the change included rg, sed, apply_patch, and git commit.

Testing

• No automated tests were run for this change.

---

Codex Task