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

Definition
- z = x * scale + bias
- m = sinc(z) = sin(z)/z (with stable small-z handling)
- g = sigmoid(alpha * m + beta)
- y = x * g

Goal
- Fuse operations to reduce memory passes; target ≥1.30x speedup with accurate small-z handling.
