[Breaking] Adding dilation/groups to Conv2D. Adding dilation to Pool2D

[coreylowman]

Resolves #693
Resolves #589

Breaking API changes

• Two new parameters are added to Conv2D and the Min/Max/Avg Pool2D nn layers
• API for tensor op conv2d is now implemented on a tuple of the image & weight, instead of accepting filters as a parameter. Additionally, the stride/padding/dilations/groups are now passed into the method instead of being const parameters. A call to conv2d with default parameters for things would look like this:

(img, filters).conv2d(Const::<1>, Const<0>, Const::<1>, Const::<1>);

• Removed min_pool2d/max_pool2d/avg_pool2d
• TryPool2D now accepts the type of pooling/kernel/stride/padding/dilation as parameters:

img.pool2d(Pool2DKind::Min, Const::<2>, Const::<1>, Const::<0>, Const::<1>);

New features

• Added dilation parameter to conv2d
• Added group parameter to conv2d
• Added dilation parameter to pool2d
• All dimensions of image/filter tensors support usize now (previously channel was const)
• Stride/padding/dilation/group can now be usize values if the tensor dimensions are usize

Tests ran

• cargo +nightly test --tests
• cargo +nightly test --tests -F test-f64
• cargo +nightly test --tests -F test-f16
• cargo +nightly test --tests -F cuda
• cargo +nightly test --tests -F cuda,test-f64
• cargo +nightly test --tests -F cuda,test-f16
• cargo +nightly test --tests -F cudnn
• cargo +nightly test --tests -F cudnn,test-f64
• cargo +nightly test --tests -F cudnn,test-f16