Decouple output scheduling from solver state

[AhmedSalih3d]

Motivation

• Output boundaries reset local integration state which caused neighbor-list rebuilds and in-place particle sorting to be triggered by OutputTimes, changing physics outcomes.
• The solver must keep adaptive timestep and neighbor-rebuild displacement independent of the passive output scheduler so snapshots do not affect solver sequencing.

Description

• Added a persistent runtime state SimulationRuntimeState holding Δx, dt, and an initialization flag, and threaded it into SimulationLoop to avoid resetting these when the loop is re-entered (src/SPHCellList.jl).
• Changed SimulationLoop signature to accept Runtime::SimulationRuntimeState{FloatType} and updated the loop to read/write Runtime.Δx and Runtime.dt, and to perform the initial neighbor-list build only once.
• Initialized the Runtime instance in RunSimulation and passed it into SimulationLoop so neighbor-list rebuild timing and adaptive dt persist across outputs (src/SPHCellList.jl).
• Fixed next_output_time(times::AbstractVector, ...) indexing to use <= length(times) so the final configured output time is reachable (src/TimeStepping.jl).
• Restored an allocation-free Δt method that accepts (Position, Velocity, Acceleration, SimulationConstants, SPHKernel) and computes the max acceleration from the acceleration vector for compatibility with existing tests (src/TimeStepping.jl).

Testing

• Ran julia --project=. -e 'using SPHExample' which precompiled and loaded the package successfully.
• Ran julia --project=. -e 'using Pkg; Pkg.test()'; the time stepping tests passed, but the isolated particle test errored due to a FullTimeStep signature mismatch in the test harness (existing test calls FullTimeStep with the older 7-argument signature), so the full test suite did not complete successfully.
• The changes are confined to src/SPHCellList.jl and src/TimeStepping.jl and preserve the intended behaviour of keeping OutputTimes passive with respect to solver state.

---

Codex Task