Make PresetTimeCallback robust to tstop roundoff

[AshtonSBradley]

Summary

• Fixes PresetTimeCallback matching when an integrator reaches a preset floating-point time within roundoff instead of at the exact stored tstop.
• Keeps the exact insorted path, then checks the nearest sorted preset time with a small floating-point tolerance after promoting the solver time and preset time together.
• Adds focused regressions for a 9.0 preset time reached as 8.999999999999996, covering integer vector, Julia range, and floating vector schedules; this is the callback-side behavior behind PresetTimeCallback skipped for certain constant lags in DDEs DifferentialEquations.jl#1124.

Explanation

PresetTimeCallback currently checks preset times with exact sorted membership. That can skip the callback when upstream time stepping lands on the intended stop with a representational roundoff, which showed up in the downstream DDE reproducer from SciML/DifferentialEquations.jl#1124.

This makes the callback predicate robust to that case by preserving the exact membership check first, then comparing only the nearest candidate tstops from the sorted preset-time vector. The comparison promotes t and the candidate tstop together, so schedules like [2, 9] and 2:7:9 are handled the same way as [2.0, 9.0] when the solver time is floating-point. Non-floating comparisons and exact matches keep the existing behavior.

Tests

• julia --project=. -e 'using Pkg; Pkg.test()'
  • Passed: DiffEqCallbacks | 373 pass / 373 total.
• julia -e 'using Pkg; Pkg.activate(; temp=true); Pkg.develop(path="/private/tmp/DiffEqCallbacks-1124"); Pkg.add("OrdinaryDiffEqTsit5"); include("/private/tmp/DiffEqCallbacks-1124/test/preset_time.jl")'
  • Passed with no test failures.
• Downstream DDE reproducer based on PresetTimeCallback skipped for certain constant lags in DDEs DifferentialEquations.jl#1124 with local DiffEqCallbacks and explicit MethodOfSteps(Tsit5()).
  • Passed: fired preset events included (:up, 9.0) and sol.retcode = SciMLBase.ReturnCode.Success.
• julia --project=@runic -e 'using Runic; exit(Runic.main(["--check", "src/preset_time.jl", "test/preset_time.jl"]))'
  • Passed.
• git diff --check
  • Passed.

[ChrisRackauckas]

Why not just use the values from Julia's range?

[ChrisRackauckas]

Seems like this was AI slip, not the real issue.