[P] Weber Electrodynamic Optimizer + SDR Hardware Entropy for Autonomous ML Research (fork of karpathy/autoresearch)

We forked karpathy/autoresearch and added several things we found interesting. Weber electrodynamic optimizer. We applied Weber's force law bracket from 19th-century electrodynamics to gradient descent. The bracket W = 1 − v²/(2c²) + ·a/c² modifies the effective learning rate per-parameter based on its momentum (velocity) and change in momentum (acceleration