feat: add Conv1D layer

[ulises-jeremias]

Only Conv2D exists. Conv1D is useful for sequence/time-series data (e.g., IMDB
sentiment classification).

Task

Following the pattern of nn/layers/conv2d.v and nn/internal/conv2d.v:

1. Add nn/layers/conv1d.v — Conv1D layer (in_channels, out_channels, kernel_size, stride, padding)
2. Add nn/internal/conv1d.v — CPU forward/backward kernels (im2col 1D or direct loop)
3. Add Conv1D autograd gate

Verification

• Conv1D forward matches PyTorch (nn.Conv1d) for same weights
• Gradient check passes