Pedagogy: add interactive method comparison to 50_ode/30_Runge_Kutta with detailed explanations

Add an interactive comparison to `50_ode/30_Runge_Kutta.ipynb` that lets learners select a step size and an ODE, then plots Forward Euler, Heun (RK2), and RK4 side by side against the reference solution — with a log-error subplot showing the order-of-accuracy slopes.

## Code scope

- Slider for `h` (log scale).
- Optional: dropdown for ODE choice (linear decay, harmonic oscillator, mildly stiff).
- Three trajectories overlaid + reference.
- Log-log error vs `h` subplot showing slopes 1, 2, 4 (the orders of accuracy).

## Explanation scope (the deliverable)

- **What "order of accuracy" actually means** — error scales as `O(hᵖ)`; doubling `p` doesn't double accuracy, it *squares* the rate at which accuracy improves as `h` shrinks.
- **Per-step cost vs convergence trade-off** — RK4 costs 4 function evaluations per step but its error is `O(h⁴)`. For a target error, the total work (steps × evals) is dramatically lower than Forward Euler. This is the practical reason RK4 is the workhorse.
- **Why halving the step doesn't help equally for all methods** — shows up directly on the log-error plot as different slopes.
- **What this notebook is *not*** — adaptive step control, embedded RK pairs, dense output. Those are why production code uses `solve_ivp` (`11`); this notebook is about the underlying methods.

## Why this framing

Comparison-by-table loses the visual punch of slope difference on a log-log plot. The slider lets the learner discover the order-of-accuracy directly. The explanation is what turns "neat plot" into "I understand why we use RK4 in practice."