convex domain for ball

[ducoffeM]

• the clone method was not working for ball due to the use of ConvexTypeDomain and not convertir the name ball to the associated ConvexTypeDomain object.
• tests in test_clone are covering the different domains
• in backward_layers relu_ get_upper and get_lower were not using convex_domain

[nhuet]

Beware, this function is already defined in the same file. You should modify the name (something like test_convert_1D_ball ?)

[nhuet]

I rebased the branch on my fork so that

• I use the new perturbation domain classes instead of old ConvexDomainType
• I fix the use of get_upper/get_lower in all places where they are called to specify each time the proper perturbation domain
• I use the new fixtures in test_clone_ball
• I use the new default decimal for assert_output_properties_ball, and no more "name" arg (o be consistent with other assert functions)
• I use an appropriate radius and ball center (it seems to me that the previous radius was actually the diameter).

But now the test is failing, either because i failed using appropriate inputs for the decomon/keras models, or because the bounds computation for ball domain has issues. (Probably the first point :D)

You can see this work on my branch convex_domain_ball on my fork.