Support collections of tensors as inputs to compiled functions

[pranavm-nvidia]

Our compiled functions currently only support Tensor types as runtime inputs. We should allow collections of Tensors and do whatever unpacking is necessary in the executable code.

[akhilg-nv]

#701