[bh-official]

Closes #97

What

The EKF localizer only had labels like # predict state and # kalman gain. This adds the equations next to the code that implements them, so a beginner can match each step to an equation.

Changes (comments and docstrings only, no behavior change)

• Class docstring: notation, predict/update equations, references (Thrun, Burgard, Fox, "Probabilistic Robotics", chapter 3; Kalman 1960; PythonRobotics).
• Constructor docstring: all noise arguments with units (standard deviations, m/s^2, deg/s, m).
• update(): equation comments for state prediction, covariance prediction, predicted observation, innovation covariance, Kalman gain, state and covariance update. Notes that P - K S K^T equals (I - K H) P.
• _jacobian_F, _jacobian_G, JACOB_H, _observation_model: where each Jacobian comes from (derivatives of State.motion_model).
• Documents two things the code does that differ from the textbook, without changing them: the input noise is scaled by the variance matrix instead of the standard deviation, and F and G are evaluated at the predicted state instead of the last state. If you prefer to change either, I can do that in a separate PR.
• Fixed the typo "extimated".

Checked

pytest -q test/test_ekf_localization_path_tracking.py test/test_ekf_vs_ukf_comparison.py test/test_ekf_ukf_pf_comparison.py
3 passed (before and after)

git diff shows only comment/docstring lines added, plus the typo fix in one docstring line.