Change binops for-each kernel to thrust::for_each_n

[davidwendt]

Description

Replaces the custom for_each_kernel in binary_ops.cuh with thrust::for_each_n

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[davidwendt]

Binary-ops benchmark runs comparison for RTX A6000
binops-benchmarks.txt

There are some interesting regressions but also some significant speedups.

[davidwendt]

Using exec_policy_nosync shows much better results overall.
binops-benchmarks2.txt

[davidwendt]

/merge

[karthikeyann]

Interesting!
NVIDIA/cccl#1302 PR introduced cub::DeviceFor::ForEachN and further improvements to thrust::for_each. It's a good idea to consider replacing all other custom for_each like kernels with thrust::for_each