Operator: PseudoHuber-Affine-Gate (Fused CUDA Kernel)

Definition
- z = x * scale + bias
- m = delta^2 * (sqrt(1 + (z/delta)^2) - 1), with delta=1
- g = sigmoid(alpha * m + beta)
- y = x * g

Goal
- Elementwise fusion with smooth robust activation; target ≥1.30x speedup.
