Function conv 

pub fn conv<D: Float>(
    c_in: usize,
    c_out: usize,
    k: usize,
    s: usize,
) -> impl Fn(SymTensor) -> SymTensor

Conv2d → BatchNorm2d → SiLU — the basic building block of YOLO26.

Matches ultralytics.nn.modules.conv.Conv(c1, c2, k, s):

• bias=False (BN subsumes the bias term)
• autopad: p = k / 2 (same-padding for odd kernels)
• BN eps=0.001 matching the pretrained YOLO26n weights (trained with eps=1e-3)