pytorch/pytorch#54138

In Python, implementing type resolution across Windows
and Linux/macOS is tricky, as the same `l` specifier can
map to both `int32` and `int64`. Moreover, expanding to
to 8-byte integers the `u8` specifiers can mean both
8-bit and 8-byte integers, causing confusion.

The same issue is discussed in detail in the PyBind11 repo:
pybind/pybind11#1908

This commit changes the low-level naming convention
to make it more friendly to upcoming matrix-multiplication
kernels.