Conv2d general

[M1ngXU]

#770

[M1ngXU]

only for cpu rn, i'll try to impl cudnn (and maybe cuda)

[M1ngXU]

cudnn should be easy, no idea about cuda though. also, how should i test it? i currently only test with the old tests, though this won't show errors if i someplace swapped x and y.

[coreylowman]

Fyi @M1ngXU I just merged some pretty breaking changes for conv that you'll want to start with. It should be easier to work with hopefully

[M1ngXU]

i'll drop this pr for now and just use it locally; maybe someone else can implement this. This should probably add non-squared stuff for pool/upsample/... too.