by allowing min_size_per_dim to be longer than the number of dims in a tensor, it makes usage easier.
for example,
min_size_per_dim=(10_000, -1)
would then work for tensors of single dim - e.g. of size (1234, )
and also for tensors of 2(+) dims - e.g. of size (4321, 100)
this is much more nice for the user in several cases.
by allowing min_size_per_dim to be longer than the number of dims in a tensor, it makes usage easier.
for example,
min_size_per_dim=(10_000, -1) would then work for tensors of single dim - e.g. of size (1234, ) and also for tensors of 2(+) dims - e.g. of size (4321, 100)
this is much more nice for the user in several cases.
also - few changes for scalars inputs+outputs