[FEA] Replace the use of `cudf::detail::copy_if_safe` with the new `cudf::detail::copy_if`

## Description

With the new `cudf::detail::copy_if` added to `algorithm.cuh` in #20861 which uses CUB and can handle more than 2B inputs at a time, we don't need the `safe` version anymore. See: https://github.com/NVIDIA/cccl/issues/1584 and https://github.com/NVIDIA/cccl/pull/1612.

PR #20861 replaces one of the two overloads of the `cudf::detail::copy_if_safe` with `cudf::detail::copy_if` but the other one (the one with stencil) needs to be implemented and replaced. Related issue https://github.com/NVIDIA/spark-rapids-jni/issues/4078

_Originally posted by @mhaseeb123 in [#20861](https://github.com/rapidsai/cudf/pull/20861/changes#r2621523599)_