Update example_device_radix_sort.cu

[eriktedhamre]

Description

Updates the Pair::operator< in example_device_radix_sort.cu to treat +0.f and -0.f as equivalent.

closes Issue 1575

Checklist

I assume adding tests is irrelevant due to this being an example. I did test that my solution produces the desired result.

Something to note is that the how to compile example: nvcc -arch=sm_80 example_device_radix_sort.cu -I../.. -lcudart -O3 did not work for me when standing in cccl/cub/examples/device.
But produced: In file included from ../../cub/util_allocator.cuh:36, from example_device_radix_sort.cu:45: ../../cub/config.cuh:36:10: fatal error: cuda/__cccl_config: No such file or directory 36 | #include <cuda/__cccl_config> | ^~~~~~~~~~~~~~~~~~~~ compilation terminated.
I could however run nvcc -arch=sm_80 examples/device/example_device_radix_sort.cu -Iexamples/ -lcudart from cccl/cub.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[alliepiper]

It looks like we can simplify this further to just:

bool operator<(const Pair& b) const
{
  return key < b.key;
}

The default rules for float comparison will give us what we want regarding +/- zeros: https://godbolt.org/z/jds6nWfEh

[alliepiper]

As for building them, the instructions are likely out of date and we typically just use CMake for everything.

We can make build examples easier by adding some CMake presets and updating the documentation to show how to use them. I've filed a feature request for this, #1617.

[alliepiper]

/ok to test

[alliepiper]

LGTM -- thanks for the contribution!