Refactor dynamics parameters to use numba typed dict

[dhruvildarji]

Summary

• Replace individual float parameters (16+ args) in vehicle dynamics functions (vehicle_dynamics_ks, vehicle_dynamics_st, accl_constraints, steering_constraint) with a single numba.typed.Dict parameter
• Add create_numba_params() helper to convert plain Python dicts to numba typed dicts
• Simplify call sites in base_classes.py by passing self.numba_params instead of unpacking 16 individual self.params[...] values per call (5 call sites in RK4 + Euler integrator)
• All existing unit tests pass with identical numerical results

Motivation

As noted in #157, the current approach of passing 16+ individual float parameters through every dynamics function call is verbose and error-prone. Using a numba typed dict:

• Reduces function signature complexity from 18 args to 3 args
• Makes it trivial to add new vehicle parameters without changing every function signature and call site
• Maintains numba JIT compilation compatibility and performance (450k+ fps on dynamics step)

Test plan

• All DynamicsTest unit tests pass (derivatives, zero-init roll/dec/acc/rollleft)
• Performance benchmark: 450k+ fps (well above the 5000 fps threshold)
• Numerical results match ground truth values exactly

Closes #157