Stability & Generalisation of Gradient Descent for Shallow Neural Networks without the Neural Tangent Kernel