Moved kernel creation to name-specific functions.

[chase-dwelle]

Per @thuydotm , simplified some kernel creation.

• Removed create_kernel(), now kernels are created by calling their name-specific functions. These are only circle_kernel() and annulus_kernel() at the moment. Custom kernels are any numpy array and therefore don't have a function, e.g., kernel = np.ones((3, 3)).
• Removed _validate_kernel_shape() and replaced with validate_kernel(), which checks that the kernel is a numpy array and has odd dimensions. This isn't a strict requirement, but I think it makes sense to have the focal point be at the center of a kernel. Perhaps in the future we can accept custom kernels where the focal point is not centered and then pad the kernel in order to center it.
• Updated tests to reflect these changes.

[brendancol]

Hell yeah! @thuydotm let me know what you think

[thuydotm]

• Removed create_kernel(), now kernels are created by calling their name-specific functions. These are only circle_kernel() and annulus_kernel() at the moment.

circle_kernel() and annulus_kernel() look good to me.

• Custom kernels are any numpy array and therefore don't have a function, e.g., kernel = np.ones((3, 3)).
• Removed _validate_kernel_shape() and replaced with validate_kernel(), which checks that the kernel is a numpy array and has odd dimensions.

@chase-dwelle We should have a function name custom_kernel which accepts a numpy array as an argument (we may consider other array-like objects as long as we can validate them). The logic in validate_kernel() should be inside custom_kernel also. For now, the function either raises an error or returns the array itself.

[chase-dwelle]

@chase-dwelle We should have a function name custom_kernel which accepts a numpy array as an argument (we may consider other array-like objects as long as we can validate them). The logic in validate_kernel() should be inside custom_kernel also. For now, the function either raises an error or returns the array itself.

Changed. Agree on accepting any array-like object, but need to figure out if it would break any downstream dependencies, though assume we could just convert it to a numpy array during validation.

[thuydotm]

@chase-dwelle the return of custom_kernel needs to be a numpy array for sure.

@brendancol this PR looks good and ready to be merged.