Adding Support for Lazy Linear Layers

[sgstepaniants]

Just wanted to ask if at any point, there is consideration to add support for Lazy Linear layers like in PyTorch: https://pytorch.org/docs/stable/generated/torch.nn.LazyLinear.html

This is a really useful feature, especially for convolution network architectures.

[lockwo]

Relevant previous discussion: https://github.com/patrick-kidger/equinox/issues/949