Branch shifting limiter terms

[AhmedSalih3d]

Motivation

• Avoid computing shift-related vector expressions for non-fluid interaction pairs because boolean-multiplied vector expressions can still perform expensive math even when the logical condition is false, so a branch can save work when many neighbours are boundary/non-fluid.
• Microbenchmarking indicated the branched form is faster in a mixed fluid/boundary workload while producing identical results.

Description

• Replace MotionLimiterCondition boolean scalar multiplication with an explicit if ParticleType[i] == Fluid && ParticleType[j] == Fluid branch in the kernel-output shifting path in src/SPHCellList.jl so Vᵢ and shift accumulations are only computed for fluid-fluid pairs.
• Apply the same explicit fluid-fluid branch in the no-kernel-output shifting path in src/SPHCellList.jl to avoid computing unnecessary shift_c_acc and shift_r_acc terms for non-fluid pairs.
• Commit contains only these branch replacements and associated minor reformatting in src/SPHCellList.jl.

Testing

• Ran julia --project=. -e 'using Pkg; Pkg.test()', which errored in an existing time stepping test due to a preexisting MethodError: no method matching Δt(...) unrelated to this change (tests did not pass).
• Ran julia --project=. -e 'using SPHExample; println("loaded")' which succeeded and precompiled the package.
• Executed an ad-hoc automated microbenchmark comparing boolean-multiplication vs branched accumulation over 5,000,000 mixed pairs, which returned boolmul seconds=0.016947455, branched seconds=0.006823296, and identical numeric results (same=true).
• Ran git diff --check (no whitespace/errors reported) as part of verification.

---

Codex Task