Update convert_inputs_to_tensors to require Python numbers

[pranavm-nvidia]

convert_inputs_to_tensors currently passes any non-tensor arguments directly to the Tensor constructor, which means that it ends up enabling more than just Python numbers (which was its original intention) in our APIs. The issue is that this makes the API inconsistent, with some APIs working fine with DLPack tensors and others not.

We should update convert_inputs_to_tensors to only accept Python numbers to avoid this.

This change will also make it slightly easier to fix #154 since we can just look at the Python type to determine whether or not to allow automatic type conversions.

[pranavm-nvidia]

Implemented by #185