feat(gds-sim): continuous-time ODE integration support

[rororowyourboat]

Problem

gds-sim only supports discrete-time simulation (timestep-based). Many dynamical systems — differential games, optimal control, continuous-time control theory — require continuous-time ODE solvers with adaptive step control.

Proposed Solution

Add a ContinuousModel to gds-sim wrapping scipy.solve_ivp:

• ODE right-hand side function specification
• Adaptive RK45/RK23 integration with configurable tolerances
• Backward-time integration support (for reachable sets, adjoint equations)
• Compatible with gds-psuu (Metric extraction from continuous results)

Motivation

The Homicidal Chauffeur notebook demonstrates a pursuit-evasion differential game that requires solve_ivp at rtol=1e-10. This class of problem cannot be approximated well with discrete timesteps.

Scope

• ContinuousModel class with rhs function and time span
• Integration backend using SciPy
• ContinuousResults compatible with existing Results columnar storage
• Backward integration via time reversal